9080 02246 0684 3 MIT LIBRARIES

advertisement
MIT LIBRARIES
3 9080 02246 0684
I
LIBRARIES
J
k^
DEWEY
HD28
,M414
Oi
MIT
Sloan School of Management
Sloan Working Paper 4213-01
August 2001
A SELF-CONFIGURING AND SELF-ADMINISTERING
NAME SYSTEM WITH DYNAMIC ADDRESS ASSIGNMENT
Paul Huck, Michael Butler,
Amar Gupta, Michael Feng
This paper also can be downloaded without charge from the
Social Science Research
Network Electronic Paper Collection:
http://papers.ssm.coni/paper. taf?abstract_id=289254
A
Self-Configuring and Self-Administering Name System with
Dynamic Address Assignment
Paul
Huck
Oracle Corporation
Michael Butler
MITRE Corporation
Amar Gupta
Massachusetts Institute of Technology
Michael Feng
Oracle Corporation
Abstract
A distributed system that stores name-to-address bindings and provides name resolution to a network of computers is presented in
this paper.
This
name system
najTie service consists
(DNS).
The
DNS
consists
of a network of name services
of an agent program
that
in
offers high scalability
and feult-tolerance
Name Domain (BIND)
BIND process based on the changing
capabilities
and
conjunction with the current implementation of the
agent program automatically configures a Berkeley Intemet
dynamically reconfigures and administers the
1
works
that are individually self-configuring
state
The
self-administering.
Domain Name System
process during
start
up and
of the network. The proposed name system
and communicates using standard Intemet protocols.
Introduction
As we approach
protocol suite, has
shown
the close
of their second decade of use, the underlying technology of the
a tremendous resilience that attests to the fiamers' foresight. Their achievement
one considers
that this architecture
was conceived
and
few types of digital
transactions.
relatively
Internet, specifically the
at
a time
when
there
were remarkably few network
is
TCP/IP
even more remaikable when
hosts,
very low bandwidth
Computer mobility was unheard of and corporations and campuses were
links,
generally
centralized at a single geographic location.
Today, the Intemet
is
rapidly approaching a billion hosts; link bandwidths vary
irom paltry himdred bits per second to
per second; links carry real-time voice, video, and data trafBc including financial, medical, criminal records, which place
privacy as well as performance; host mobility
virtual private
fluidity
is
commonplace, and organizations
networks. With the advent of "titemet Ready"
of the Intemet to continue.
are
now
gigabits
demands on
geographically distributed are connected by
embedded processors, we can expect these
increases in size, diversity,
and
The underlying
One such
expected as the hitemet continues to evolve.
networks
(e.g.,
home
state
all
hosts
would have unique, permanently assigned IP addresses'
into three classes.
saw
among the participating
A
fit.
few years ago
—displace IPv4 and
its
One of the ways
this
.
a conscious effort or
masquerading
it
presence of host mobility, micro
on address
tlie ftiture.
It
was envisioned
that all interfaces
Thus, large blocks of addresses were assigned to an
institution or a
of the allocated block was determined by the organizations expected needs and
size
them both
it
distributed the
the responsibility
management of address space
inefficient
was being
management of the
and authority
for
fell
four billion available
managing a portion of the address
rapidly exhausted
One of the
lead to a short-term
crisis;
it
appeared
that the then
solutions, IPv6, offered a presently "inconceivably large"
Although several IPv6 implementations are
now
available,
IPv6 has not—and
in feet
might
paltry address space.
that the "address
desire tor a globally unique,
in the
were wastefiil^, the perception was that the available 32-bit address space was inconceivably large.
128-bit address space as an alternative.
DHCP,
The
organizations giving
"inconceivably large" address space
was
for
management
the discussion with a brief historical perspective
A, B, and C. This approach worked exceptionally well since
space. Although such allocations
never
We begin
as ever, but weaknesses (growing pains?) are to be
of the hitemet, address space management was deceptively simple.
corporation to administer as they
addresses
area involves address space
of things today, and explore an alternative
In the infancy stages
now
sound
Internet are as
networks), and network reconfiguration.
management, recap the
on
of the
architectural concepts
space
crisis" in
IPv4 was averted was the rethinking of one of IP's
permanent address assignment for every host
occurred accidentally, but address space
firewalls,
and VPNs.
interface.
It is
initial
assumptions—the
arguable whether this change in assumptions
management changed
fijndamentally with the introduction of
We now routinely violate the permanency and even the uniqueness of address assignment
by recycling IP addresses both temporally and spatially.
Such
of a connection
loss
of a
Q^ansient assignments have,
is
DHCP
work around
for
integrally
lease (e.g.,
by a dropped PPP connection)
on the changes
Excepting experimental subnets
A
effectively severs
TCP's dependence on address permanence.
packets are transiently tunneled to the "care
Class
artifacts in
other protocols; specifically, TCP's concept
coupled to the invariance of source and destination address as presented in the packet header. Hence, the
providing routing changes based
A
however, exposed some undesirable
in
in
It
offers
any open
TCP the
illusion
TCP connections.
Mobile-IP
1
in large part,
a
of a permanent IP address while simultaneously
IP address. (With MobilelP, Mobile hosts retain their permanent addresses but their
of address of a mobile agent at their point of attachment to the network.)
the
is,
92. 1 68. x.y address space
address allocation contains 16 million host addresses
Dynamic address assignment
is,
however, an exceedingly powerful technique
address space problem, but the larger address space
example, could be optimized
dynamic address assignment
if
management and network
to the logical conclusion
making
concept of trunking, and thereby avoiding the address space problem
One might
ask
why
conplementary and address
fixed-size, short
space
is
the concepts
of
different problems.
explicitly supporting the
and host names are both preserved
Internet addresses, the 32-bit values that
tiierefore, prescribe its
appear
in
The two techniques
of the address, which greatly reduces the
bits
compact and very convenient
for
size
of routing tables
in
are
IP packets, provide a convenient,
migration through
tiie
When the address
network.
carefijUy administered, addresses also provide sufficient information for route aggregation or "trunking", routing traffic based
only a few
on
both hosts and routers. This shorthand representation
is
machine processing.
However, numeric addresses, whether permanently or
for
and
in its entirety.
interface addresses
hand indication of a packet's destination and,
shrinking
(Asynchronous Transfer Mode) protocols took such
interfece addresses locally assigned
all
tlie
configuration problems as well. Route aggregatioa for
ATM
address reassignment were possible.
only
that offers solutions to not
transiently assigned, are
an exceptionally inconvenient representation
humans. Dynamically assigned addresses only exacerbate the problem Host names provide much better mnemonics—as well as a
much
larger total address space
and
tlie
opportunity for many-to-many name-to-address mappings as well as dynamic address
assignment
Presently in the Internet,
Mobile
IP,
and IP masquerading
is
dynamic address assignment has been applied primarily to end-user machines using either
used in firewalls
at the enterprise
boundaries. Virtual servers (where a single virtual
of several physical hosts) offer another example of dynamic address binding, though the implementation
such uses have shown the power of dynamic address binding,
management in
dynamic address assignment The
mnemonic name space and
address space;
still
do not
exploit
URL maps to one
differs in details.
dynamic address binding
Although
to obviate address
networt: design.
Setendipitously, the early inclusion
for
we
DHCP or
of this mnemonic "cmtch" which allowed named hosts has also provided
hierarchically organized
the smaller
tiie
cornerstone
DNS servers universally accomplish the mapping between the large and
and compact addresses space,
b
order to reap the benefits of a
more dynamically assigned
we will need a more fluid name-to-address mapping than in presently available.
This paper addresses the key issues associated with modifying
DNS to enable more dynamic address assignment.
2 Overview
The Autonomous Network Management
attenpts to develop a collection of protocols and software tools
initiative
enable networks to be rapidly configured and maintained by untrained personnel. Industry efforts such as
ZeroConf Working Group have attempted
on providing
a solution for endhosts
planning and
management
to address
some
issues related to
problem
into four inter-related subtasks: address assignment; host
augmentation; and resource redistribution for load balancing. In
providing dynamic interface address assignment for
renegotiate addresses each time the network
scheme must be
is
network
all
extended.
To
implementations [28]. Nearly
In fact,
specification
it is
BIND
In particular,
intervention after
focus on the
first
two
to
automate the network
name
aspects.
registration;
Our approach
network
focuses
on
so that the network components can themselves efficiently
facilitate the latter,
a more flexible
DNS
it
has
Name Domain (BIND)
hostname
to address bitiding
new DNS
it
interoperability
and thus permit and encourage
desirable if a
TTie current implementation
added or removed fi-om the network, these
files tell
files
BIND the zone
turned on and
or as
network boundanes.
In
In a
new
this
architecture.
The
DNS implementation.
DNS
Thus,
responsible
addition, if the
significant
This
it is
for,
is
for user
to write several configuration
the location
of other DNS servers on
configuration can be a tedious process, as every time a
must be updated. In
DNS zones and ownership of the zones may have changed too.
clients.
each
of BIND requires a network administrator
removing name servers or by merging two networks togetha),
mobile or wireless
flexibility in
DNS
DNS service could be run with no lengthy configuration process and no need
and the location of the root name server Setting up
off,
iinplementation or commercial products derived fi'om
implementations with added features, while remaining compliant with existing standards.
would be
started.
DNS
has been the de facto standard for
has become virtually synonymous with the industry standards that define the
order to fijnction correctly. These configuration
the network,
assets,
we
modem networks use the public domain BIND
all
was written to allow product
possible to develop
files in
paper,
and the lETF's
incorporated.
For more than a decade, the Berkeley Internet
BIND.
this
will
automated configuratioa but these are focused exclusively
and they ignore other network conponents. TTie approach taken by us
task splits the
DHCP
tliat
new
host
is
network topology changes (such as by adding or
changes are required to the configuration of BIND as the
certainly not ideal for
home networks or networks consisting of
home-networked environment, hosts are frequently leaving and entering the network as appliances are
hosts are added.
Mobile
clients also will join
and leave networks often as they are transported across
an ideal environment, hosts should be &ee to join and leave the network, and should immediately gain access to
the services; this requires that
tiie
DNS be able to configure itself on a dynamic basis.
DNS
3
TTiis section describes the hierarchical
naming system and
examines the operations involved for performing updates
DNS was developed in
the network [22].
files
was created
It
1
resided on every host in the networic, and
become a woridwide network,
the process
ASCI]
if a
DNS
of the current
standard.
It
also
to the network.
984 by Paul Mockapetris
to replace the
distributed architecture
host
as a distributed database that could resolve the address
files that
name was not
associated host
names with
listed in the file, that
of maintaining the host
files
became
of any computer on
their respective IP addresses.
host could not be reached
As
These host
the internet
grew
to
increasingly unwieldy, leading to a growing need for the
DNS.
The
identified
the
DNS
and located
directory can
in the tree
domain name from
riglit
to
by
left,
be thought of as a
its
tree,
with each node on the tree representing a "domain"
domain name, which uses
we can
follow the path
down
one of many top-level domains; others include "com", "gov",
is
the sequence
The
the familiar dotted notation
the
"org",
(www.mitedu,
flill
Each domain
for example.).
DNS tree to the correct node (See Figure
and "uk". The
[1],
I ).
is
As we read
The "edu" domain
and unique domain name of any node
is
in the tree
of labels on the path from that node to the root
liierarchical tree
associated with
it
of domain names can be
These host data are defined
referred to as the
in resource records.
domain name
There are many
space. Eiach
domain
different types
common being the address-record (A-record), which associates the domain name with an IP address.
in the
space has host data
of resource
records, the
most
Figure
The
of one or more
DNS
domains
autiiority
over a zone
particular
DNS server.
3.1
Name
Each zone
Authoritative
name on the
the
zone that
when
owns a domaia Each
tliat
share the
the network
Therefore, the
same
DNS
DNS
Hierarchy
architecture
is its
distributed
implementatioa
DNS
servers
may be
DNS server is given authority for one or more "zones". A zone is a collection
parent domain, along with the associated resource records.
manager responsible
for the
domain
that contains that
A DNS server receives
zone delegates the authority
to that
DNS infrastructure is distributed both geographically and administratively [28].
Servers
in the
name
domain name space has
servers are the only
Internet,
is
of the
other distinguishing characteristic
operated by any organization that
1 -
two name
servers authoritative for
a resolver can only be assured that the address
entry has time-to-live to prevent staleness
server
is
A secondary name server
is
2).
is
correct if the
a primary one and a secondary one.
answer comes from an authoritative name server for
To improve performance, other name servers may cache resource records, but each cached
A name server stores all the resource records for the zone for which
of data
an authoritative server for its zone(s) and reads
to the zone's resource records, they
it,
name servers guaranteed to contain the correct resource records for their zone. When querying a
being queried (See Figure
The primary name
at least
must be made
also authoritative for
to the primary's
its
zone
zone(s); however,
it
its
file.
zone data from frie zone
This data
obtains
its
is
files [ 1 ]
sent to secondary
set the
two or more different zones.
period of this interrogatioa
is authoritative.
When changes are made
name
servers
upon request
zone data via data transfers from other name
A secondary name server interrogates the primary or other secondary servers periodically to determine if
network administrator can
.
it
its
servers.
zone's data have changed.
A
A single name server can act as both the primary and secondary server for
Query for address
www.mit.edu
Local
Name
name
Referral to .ecfu
Name
of
server
server
Server
Query for address
www.mit.edu
of
^
i
Name
Referral to mit.edu
name
server
server
Query
for
address
of w\Nw. mit.edu
Name
server
Address
S
Figure 2
3.2
-
The
original
DNS
RFC
2136
DNS specification
sent to the primary
name
(from |12|)
written with static networks in
DNS
it
can use the
It
was assumed
changing networks, several extensions to
Update protocol
server,
It
that allows for
can be sent by a
which receives the
dynamic updates
that hosts
would join and leave
notifi' signal to
inform
its
DNS
to the
have been implemented
DNS.
A
dynamic update
DNS client, DNS server, or a DHCP server (see Section 4).
signal
and permanently updates
primary server directly or passed through one or more secondary servers
update request
mind
the
modem networks, computers are fiw to join and leave the network, and many new devices are being
in
addition of resource records to a zone.
is
was
In order to deal with dynamically
[34] defines the
DNS
and Zone Transfers
network infrequently. However,
connected.
mit
www mit.edu
^
Resolution of www.mit.edu through
Dynamic
Stanford
of
secondary servers
until
it
that the
its
zone
file.
reaches the primary.
The
signal
Specifically,
is
a deletion or
The
iipdaie signal
may be
sent to the
When a primary server fiJfills
zone information has changed.
an
.
In order for multiple
zone
transfer
when
all
is
the transfer
name
servers to maintain consistency of their records, zone transfers are performed
of resource records from a primary name serva
resource records are sent
inaemental zone
Instead
of sending
all
the resource records,
an incremental zone transfer and a primary server chooses whether
fiill
zone
TTie following
authoritative
1
is
transfers
editing the
pass
it
to
a brief example of
file,
3.
will
how
the current version
is illustrated in
Figure
primary server
The primary name
if it
knows
its
A
server.
possible for the primary
last
zone
perform a
of
name
BIND
zone
flill
name
transfer or
at least
fiiU
zone
server to perfonn
an incremental one.
once every 24 hours
[3].
all
3.
of a host This
may come from
is
administrator manually
received at a secondary server,
is
continued
until the
it
will
primary server
Once the zone file is changed, the serial number of the file is incremented
server reads the edited zone
file.
The
fiiequency at
which the server rereads
its
zone
file
and checks
for
a configurable parameter of BIND.
is
The primary
server will send a notify message to
all
known secondary
secondary server(s) support(s) the notify signal, a zone transfer
will discard the notify and wait until the next scheduled
5.
an
It is
propagates changes to the resource records to
locatioa or to another secondary server. This
zone changes
If the
A
transfer occurs
servers.
The primary
server will wait
some time
between sending each notifi.' to reduce the chance of multiple secondary servers requesting zone transfers at the same
4.
basis.
A secondary name server requests
transfer.
or through an update message received. If the update message
receives the update signal.
2.
it
server receives a request for the addition or deletion
zone
tfie
is
it
be performed no more than once every 1 5 minutes and
name servers. The process
The primary
a secondary
This will transfer only those records updated since the
transfer.
recommended that
to
on a periodic
The secondary serva then
notifies
level transfa is continued until all
is
immediately
initiated.
time.
Otherwise, the secondary server
zone transfer time.
any other secondary servers
that
may be
dependent on
secondary servers have received the changed records.
it
for
zone
ttansfers.
This multi-
Secondary
DNS
Figure 3
While the dynamic update system
Each name server
administrator free.
secondary
DNS servers for
its
network, either die primary
name serva
added
so
tiiat
in its place,
primary DNS
zone.
DNS
die primary
may
in tine
-
lessen the
BIND Updates
Server
(from [28|)
amount of administrative work
network must be configured
witii tiie
for the
DNS
DNS
must be configured
server can receive updates propaly.
server. In addition, the current
extensive knowledge and
specific primary
eflfort
The above problems
does not make them
it
DNS
secondary
to recognize
Also, if tiie primary
name
server
presence of the
ttie
name
server
is
server or other
is
added
new
to the
secondary
removed and another
tiiat it is
informed of tiie
DNS system requires extensive initialization effort to ensure proper operation.
and secondary servers and a clear domain hierarchy must be defined. These
new
Zones must
tasks all require
fiom an administrator.
are addressed
monitoring BIND and dynamic
by
tiie
system described
in the
next section.
It
uses a combination of an agent program
DNS protocols to maintain a tme self-configuring and self-administrating naming system.
4 Self-Configuring and Self-Administering
Name System
Network Overview
The naming system described
company codenamed
hosts
If a
an administrator mtist manually change the configuration of every secondary server so
be properly allocated with
4.1
servers,
address of either the primary
Name servers are not tree to join and leave the network.
server or otiier secondary
name
on
die
in this section
as Research Corporation in
network This
is
done widi
virtually
tiiis
was conceived
as part of a larger network system that
document The proposed design provides
a
fiill
is
under development
array of network services to
no manual configuration and no administration. The network design allows
at
a
tiie
for a groiqj
9
of computers
be physically connected together (through Ethernet or other network media) and
to
properly configured in a working network.
Figure
tlie
Managers form
4).
name
the
On
this
be
network there will be two types of nodes: managers and hosts (See Figure 3 and
backbone of the network and provide
to address resolution service, the
after a short time, they will all
services,
such as
name
to address resolution, to the hosts.
managers also provide dynamic IP configuration
new
for
hosts, discovery
Besides
mechanisms
for
new managers, packet routing, and a database of all hosts in the system. The services directly related to the naming service are automatic
IP configuration and
manager discovery,
network system
allow any conputer, either a host or a manager, the
to
is
external administration.
The
with either type of change
gracefiiUy with the
in
as these will
of
services that the
name system will
directly interact with.
fiwdom to join or leave the network,
new node
networi< should be able to detect the presence of a
an appropriate manner, hi addition, the network system, and
without
The
flie
goal of the
need
for
any
or the absence of an existing one, and deal
in particular the
name
system, should deal
merging of two networks. Conflicts should be detected and resolved quickly.
In order to provide the self-configuring
three types
be the
and
services: the IP configuration service, tiie
Manager
name
service,
each manager in the system
manager discovery
service,
and the agent based name
self-administrating
Manager
1
Name
Service
is
designed to run
service.
The IP
2
<^=^
Name
Service
7\
7\
—
<
J^^fcsauT
iz
iz
Manager 4
Manager
Name
Service
3
<^=o
Name
Service
M
Figure 4
-
Sample Network
10
configuration service
protocol specified
is
by
handled by a standard implementation of tlie dynamic host configuration protocol (DHCP).
the
IETF
with configuration parameters for their IP interface
that provides hosts
domain name, a subnet mask, a
default gateway,
and a location of
Manager
DNS
server.
DHCP
[6].
DHCP is a network
This includes an IP address, a
also allows a host to retain a previously
1
Manager 2
Primary
Internal
Secondary
Root
Root
Internal
Server
Server
Managers
Domain
Manager 4
Domain 2
1
Primary
Server
Primary
Server
Figure 5-Name Space View of Systems shown in Figure 4
configured domain
name
while receiving an IP address. After receiving the necessary information finm a
unconfigured host's IP interface has
requires
no
all
the necessary parameters in order to begin transmitting
a host locates a
prior configuration; as
operating systems and even most
DHCP server, a previously
and receiving on the network.
DHCP server by broadcasting discover messages on the local network.
embedded devices support DHCP.
Overall,
DHCP
All
DHCP
modem
meets the goals and design objectives of the
desired self-configuring and self-administrating network.
The process of manager discovery
packets" to each interface.
other
is
accomplished by having the manager discovery process
These packets contain the source address and a unique network number
managCT discovery processes will receive the packet, and
the
network number
the
network number for the
managers
The
that
is
unknown,
new
that
manager will respond
if the
network number
to the source
of the
is
known,
p)eriodically broadcast "discover
manager
that the
the packet will
resides on.
be dropped
]f,
All
however,
"discovei^' with a "discover reply". This reply includes
manager. This allows the manager discovery process to inform the local
name
service
of any
new
appear on the network.
design and the operation of the
name service are discussed
in the next subsection.
11
4.2
Name
Service Interface
The name
service runs
on every manager
system and consists of two concurrently running processes. The
in the
first is
the
BIND implementation of DNS. As stated before, BIND is currently used in the overwhelming majority of name servers on the Internet.
It
provides a scalable, stable, and fault-tolerant basis for the
conditions and configures
BIND
automatically.
name
service.
The second process
It
two
acts as a filter for the
configuration process.
different formats
DNS
communication between
details
XNS
of both
is
reacts to
of messages.
network
BIND,
The agent program uses Berkeley
the
UDP
On port 53, the standard DNS port the agent program listens for DNS messages.
messages, sending queries directly to the
BIND
process,
and processing update messages from the IP
On port 54, the agent program listens for any agent messages coming fi'om either the manager discovery service,
or other agents in the system
because
an agent program that
Figure 6 shows the relationship and manner of communication between
agent program, the other local manager processes, as well as other managers in the network.
sockets to listen for
is
The agent messages allow an agent process
agents.
The IETF has designated
uncommon on modem servers.
port
54
to
to gain
be used
for
knowledge of other agents and
XNS
If this service is needed, another port
Clearinghouse [27].
offer a
method of
We chose to use port 54
may be specified for agent communication. The
DNS and agent message processing are discussed in following sections.
12
Agent Program
Pnrt
ifc;-^
Manager
Name
Manager
Service
AgenI Message
# 54
Port
Discovery
Service
Agent
Program
IP
Configuration
Service
BIND
^
DNS Message
(update)
Process
DNS
Host Resolver
\dNS
Agent Program
Pnrt UR'^
Figure 6
-
Relationship between agent,
|12|)
Internal Root Servers
4.3
The name
system.
servers in the system are only authoritative for the IP addresses
If the networi;
servers. Certain
serviced
name on
name
by
name
must be queried. With
services in the
the self-configuring
manager network
naming system
this in
network's
knows
name
sub-domain
the location
TTiese are not tnie
least
one
to
mediamitedu). This host
know
will
the IP address
query any local
internal root server
of a computer
name
is
is
name
and are
authoritative for the entire
the self-configuring
service will
DNS
implemented using the idea of internal root
domain
root servers as they are not authoritative for every
domain serviced by
mitedu domain
be able
naming
service.
to provide
name
that is
domain
So, as long as a
resolution for the
using the self-configuring naming system, a host in the Ics.mitedu
in another
sub-domain serviced by the naming system (such as
system, and since the manager
is
not authoritative for the mediamitedu domain,
of the mit.edu domain. This server may resolve any domain name
the resolver to the proper media.mit.edu authoritative
the Intemet root servers inay
DNS
internal root server, the
space. For example, if the entire
may wish
query the
of at
mind, the name system
act as the internal root servers
the Internet, but only for those that exist inside the
service
of the subset of hosts configured by the naming
connected to a larger network, such as the Internet or other networks serviced by the naming system,
is
servers outside the local network
will
BIND, and other processes (from
in tine
mitedu domain by
name server. For resolving names outside the domain, such as
it
directing
Intemet host names,
be queried.
13
DNS
4.4
Messages
DNS messages that the agent program processes directly are DNS updates.
The only
to the local
BIND process on port 55.
which
to the agent process,
These messages will simply be
forward the response to
in turn will
BIND process. To all hosts on the network,
DNS
4.4.1
it
appears as
if
All other messages are simply directed
DNS queries and BEND will process the query, retum an answer
tlie inquirer.
The agent process
is
the only entity with access to the
BIND is running on the normal port.
Update
DNS updates, however, reqtiire special processing by the agent DNS updates only occur when the IP configuration service
or
manager discovery
service detects the addition or deletion
then assembles the appropriate
program.
to the agent
As
forward
it
Although
stated earlier, standard
directly to the
primary master
To
[34].
try to
RR in tine zone that
primary master as
primary master
authoritative for,
this is part
receiving the update, then
Start
of Authority (SOA)
particiilar zone.
zone
exists,
it
it
is
the only server that has the definitive set
of resource records
for a particular
DNS implementations specify that any secondary server which receives a DNS update message must
if
its
The secondary servers
is
manager service
DNS updates may be sent to any DNS server that is authoritative for a zone, they will always end up
location
will
is
known, or
alternatively through the
of its
state
must search
it
will
secondary serva chain
until
it
reaches the
only learn of the update through notify signals originating fi-om the primary master.
improve the efSciency of this mechanism, the agent program
it
local
DNS update message containing the resource record that must be added, modified, or deleted and sends
being processed by the primary master for the zone, as that
zone.
of a host or manager on the network. The respective
send
it
directiy the
will
examine the update message, and
if it applies to
primary master for that zone. Each agent process has knowledge of its
information described in Section 4.6.
for the primary master for that
DNS query to the local BIND process.
The
When the
agent program
zone on the network.
SOA asks the DNS
is
To do this,
not authoritative for the zone
the agent
for the address
Upon receipt of the SOA, BINfD will perform standard DNS resolution to find the
program
will
send a
of tiie primary master
for
a
information on the zone name. If the
BIND wiU return to the agent process the IP address of the primary master for the requested zone, and the agent process will
forward the update to that address. If however, the
retum the non-existent domain error flag
authoritative for the
of zones and
is
new zone. The
(NX_DOMAIN), and
reconfigured
especially usefiol at resolving
DNS iqxiate message ^plies to a zone that does not exist on the network, BIND will
all
the agent program will configure the local
BIND process will then process the update request
BESID process
to
be
This allows for dynamic creation
naming conflicts when two networks are merged
14
4.5
Agent Messages
Agent messages are used
structure
of an agent message.
Table
The payload holds
1.
are explained in
more
Message
It
consists
the data
detail in tlie
I D;
for inter-agent
communication and as a mechanism
of a three-tield header followed by a payload section. Tlie header
from the message. The data are
Opcode ;S/R flag
Figure 7
1
- Agent Message Header Fields
Header Field
specific to
I
Header
Payload
-
Figure 7 shows the
fields are
explained
in
each Opcode. The data sent with each agent message
following sections.
Opcode specific Data
Table
for agent discovery.
Agent Message Format
Table 2
-
Opcode
Opcode Values
Primary Master for
Primary IVIasterfor
zone a.mit.edu
zone a.mit.edu
Primary Master
Primary Master
for
zone a.mit.edu
Figure 8(a)
-
for
zone a.mit.edu
Merging of two previously unconnected networks with name
conflicts begin witii
new physical
connection joining the two.
Figure 8(b)
-
Agent programs
in
both networlts receive discover messages informing them of the newly
discovered managers.
The becomeslave message can
also
be used to make an agent a slave for a zone. However, while theforceslave message
only used between two competing primary servers, the becomeslave can be sent to any agent
but only to notify agents
of new primary
becomeslave message
passed to
is
all
servers.
the old slaves
becomeslave message informs them of the
The getroot message
agent
is
is
If a
new primary
in the
system.
server for a zone leaves the network
of the former primary server and
to the slaves
It is
not used for elections,
and a new one
of the
is
is
elected, the
loser in the electioa
The
new primary server and allows the agents to reconfigure themselves accordingly.
used by agents to share the
introduced to the system and wishes to
know
internal root server information.
the internal root server.
The
This
is usefiil
when
a
new unconfigured
getroot fimction operates on the basis that any
17
configured
DNS
server that
is
queried with the getroot message will have a reference to a root server,
server or the root server itself TTiis assumption
is
valid because the server
\n the tbnner case, the authoritative root server will simply return
have a root server reference
that will
Figure 8(c)
-
The
for
a.
Figure 8(d)
for
information to the querying server. In the other case,
Primary Master
it
will
for
a. mit.edu
Primary Master for
zone
mit.edu
conflicting
-
either authoritative or not authoritative for the particular root
zone
a. mit.edu
Primary Master
zone
own
can be anotha
be used to query for the relevant zone information.
Primary Master
zone
its
is
this root server
a. mit.edu
primary master servers send each other negotiatemaster messages.
The winner
of the election sends a forceslave message to the loser.
18
Primary Master
Slave
for
for
zone a.mit.edu
zone a.mit.edu
Figure 8(e) -The
new
slave performs a zone transfer with the master.
State Information
4.6
To
assist
log of all messages
with the agent tasks, each agent stores state information about
it
itself and other
has received In particular, the agent will keep a record for all managers that
server record contains the name, IP address, a uniqiie server ID, and a status flag that
(See Table
3).
Each server record
is
it
in the
network, as well as a
knows to exist on the network. This
server
is
configured or just starting up
placed into one or more of the following categories:
•
Own Servers - contains the agent program' s own server information.
•
Known Servers - list of server records for every managCT on the network.
•
Root Servers - server records
•
Newly Discovered
record will be
tells if the
managers
Serveis
for every
- list
domain root server on the network
of recently discovered managers. Once the agent processes the discover message, the server
moved to one of the above categories.
19
Table 3
-
Server Record Fields
Server Record Field
No
discovers
received
/
Configure as
Internal Roi
Own Server ID is
lower than Server
of a discovered
server /
Walt for a
Figure 9
4.7.2
have the requested name
When the agent is
in its database,
in the
it
Configured
it
is
every zone, the agent
is
ready to handle any
state,
it
listens for queries,
random
server out
will
of the
fimction
is
Servers
does not
it
is
a primary
used to find more slaves for a primary master. For
check tie zone information to see
Known
if it
down the DNS
DNS updates, discover and leave messages, and (if
The getSlave
a primary master for the getSlave fimction
slaves are needed, the agent picks a
name resolution query on the network. Even
can query the internal root server and find an answer by working recursively
master for a zone) periodically runs the getSlave functioa
more
Startup Scenario
Configured
Once an agent has entensd the Configured state,
tree.
-
list
if more slaves are
that is not already
needed. If
a slave for the zone.
It
then
sends a becomeslave message to that server.
The behavior of the agent
in response to
both
DNS queries and updates is described
leave message signals a change in network topology, as managers have either joined or
in
left
Section 4.4,
The
the network.
arrival
of a discover or
In the case
of a discover
message, there are two options: the discovery of a configured manager, or the discovery of an unconfigured inanager. The case of the
21
discovered unconfigured manager
any gettvot messages
it
is
an agent need only record the seirer record of the
rather simple as
when two
configured networks are joined. Network unions are revealed by the inanager
discovery process as two managers will have different network IDs that were previously unknown to
possible to have a conflict
where two managers
are primary masters for the
same
is
initiated
when
message to the newly discovered server with highest ID (the server ID
be used as
well).
Only primary masters need
and transfer the new network information
Zone
root server,
conflicts
it
can only occur
would be impossible
primary masters.
in
if the
for
a
fiiture
tliis
case,
it
is
mechanism
message indicating
that
then no conflict exists.
However,
it
can
still
it
is
new configured
When
an agent
is
a
managers, the agent will send a getroot
used so that only one server is contacted, any other metric could
scheme, as any
slave's
master will detect the conflict
zone transfer.
different internal root servers.
DNS
conflicts to occur as a single
If both
networks shared the same
root tree will not allow
flie
same zone
to
internal
have two
When the two internal root servers are different, an agent must make sure diat it is the only primary master for
SOA query to die differing root server for every zone that
differing root server will then respond with the address
exists
In
zone. Therefore an agent needs to have a
to participate in the conflict detection
two networks have
zone
zone. Tlierefore, the agent will send a
If,
otha manager.
a configured agent receives discover messages.
primary master for a zone and receives a discover message with information on
its
tlie
and resolve this conflict
Figure 10 iUustrates the procedure that
different
respond to
may receive. When a configured manager is discovered, the process is slightly more complex.
Configured managers are discovered
to detect
new manager and
is
not aware of the zone. If the
If the
resolve
of the manager
SOA
names
message
in that
is
an
serves as primary master
for.
The
considers to be the primary master for that zone, or an error
SOA response is an error message or if the address matches the querying agent,
zone by recursing
is
it
error, this indicates the other root server is
however, the diiiering root serva indicates
and must be resolved. The conflict
that
it
that
it
down the DNS
tree
believes another
not aware of the zone on the network.
from the top node.
managa
to
be the primary master
for the zone, a conflict
resolved by the negotiatanaster master message described in Section 4.5.
22
No
No
discovers received
getroot response
within timeout
/
Resend getroot
Discover from
least one
Configured
at
configured server
Received getroot
response /
/
Root
getroot
response
'Configured^
Compare
Wait on
Compare
Servers
root
servers
Discover received from
unconfigured server
Root servers are the same
Root Servers are
different
server
database
A
a slave or master for that zone.
is
file
for
each zone
can be sent to the
in the
database
namedconff\is.
named process.
This signal
tells
The agent program controls BEND through
will use a Peri script to generate a
program
the agent
4.9
to the
the
all
name
namd process to reread
the use of the
its
information for a zone, and there will be a separate
files
are
configuration
namedconf&e and
the appropriate database
database
Once
files.
changed
files.
During
files
name
service will
SIGHUP signal
files.
start-up, the
agent program
are generated, the agent
need to handle another zone. The agent program
namedconf&e, create the appropriate database file, and send a SIGHUP
in control,
any way, a
and database
files
these
in
program
BIND is then ready to answer queries and to handle updates to the name database. Once tlie
has started, the possibility exists that the
zone information
that
tie
contains
namedcotf file or the database
If the
nconedconf&e and
will automatically start the «a?wec/ process.
namedpmcess
file
signal to the
will
append
named process. With
BIND will always be configured properly to store names and to handle all the network name resolutions.
Implementation Notes
The agent program was implemented using Gnu C++ on
that control the state
class controls the
of the agent and
network information. The main class for
this
flow of the program and handles the input and output of agent and
send and receive agent and
sockets.
store
Unix operating system. The program
the
DNS
messages, and the standard
The Agent class retrieved messages by using a
UNIX
DNS messages.
Bericeley socket library
on each socket
select call
agent program
to
check the
is
of a
Agent
class.
the
set
of classes
The Agent
UDP protocol was used to
The
was used
consists
to create the necessary
availability
UDP
of data on the socket Also,
callback flmctions were used to perform time-dependent and periodic events.
The Agent class uses a number of storage classes that store the state of the network. These storage classes
class, the
Doinains
class,
network.
The Domains
messages
that
and the
MsgLog
class.
class helps to store
The
Servers class helps to store what other managers and
what zones
the agent
is
authoritative over.
The C++
have been received by the agent program.
implementation of these storage classes. In particular, the map, the
list,
The MsgLog
standard template library
name
include the Servers
services exist in the
class stores information
on the
(STL) was used extensively
in the
and the vector template classes proved
to
be very usefiil
in storing
information.
To generate the configuration and database files
to call Perl
file
commands and
called agent.pl,
files
subroutines. All
of the
Peri
for
named, embedded Perl was used. Embedded
commands and subroutines required to
and the agent program used embedded
generate the
Peri allows a
C++ program
named files were placed
Perl to call the subroutines in agent.pl every time the
in a
named configuration
needed to be changed
24
4.10 Security Concerns
The
current implementation of the system has
no
security
mechanisms
built
mechanism
configuration-iTee and administrator-free creates a conflict, as any meaningful security
configuration [37].
The
best that
anyone can hope to accomplish
For a closed environment such as a corporate
secure.
physical security
mechanisms
to control access to the
where one can physically control access to
tiie
Ethernet,
it
is
intranet,
to
is
[32].
is
to
have none
the network. If a corporation only allows approved
to
at all,
This will work well for an environment that
is
some
of
sort
keep tlie system
simply relying on
well controlled and
machines and approved users to plug
into
impossible for an outside intruder to gain access to the network.
However, as wireless networks become increasingly popular,
[14].
require
will
minimize the configuration overhead necessary
perhaps the easiest security model
machines
Securing a network designed to be
in.
controlling physical access to networks
becomes impossible
Insecure wireless networks allow anyone in range to send and iieceive data This allows for theft of services (such as an individual
using his or her neighbor's internet connection), as well as the potential for malicious hosts to be introduced to the systera
Another reason
to
for incorporative security capabifities
another larger network, such as the Internet If the system
security
is
when the network managed by the agent-naming system
insecure,
is left
agent naming system should not open up any
insecure, relying only
new
holes into the
on redundancy and physical
throughout the network. Normal
it
world could
atterrpt to violate the network's
should be at least as secure as a standard IP network. That
DNS systera
security [35].
DNS
However, because the agent naming system allows other agents
The managers
in the agent
hosts that join the network,
fiilly
it
[19].
The second
is
DNS implementations are completely
servers are kept in a trusted, safe location
to
change
and
distributed
its
configuration,
it
is
prone to attacks from outside. If
fimctions. First they provide IP configuration information to
to address resolution service for the network.
Each of these
new
services
is
secure network, both must be secured.
There are two possible methodologies for securing the managers. The
traffic
most
the
could do significant damage.
naming system have two main
and second they provide the name
vulnerable to attack, and for a
Currently,
is,
DNS servers only allow administrators at the physical machine to change the configuration of the DNS
a malicious user could introduce a "rogue agent" into the system,
all
in the
connected
mechanisms.
When a network is managed by the agent naming system,
server.
any host
is
first is
to
use IPSec to provide network layer security for
to secure the individual protocols that provide IP configuration
and name
to address resolution
[7][16][35][36].
25
Using IPSec,
all traffic
between managers
can place a rogue manager into the network, tor
that
otlier
option
is
to utilize secure versions
DNS server authentication [10], while RFC
Authentication for
DHCP messages
IPSec; however, they
do not provide
is
3007
be able
to authenticate itself
a broad solution covering
is
When
IPSec.
IPSec
is
all
of DNS and
confidentiality
is
it
In addition, hosts
by
of data. This
desired,
is
IETF
[7].
inalicious user
came from a known managa.
DHCP. RFC 2535
the
no
can authenticate inanagers, so
defines
DNS extensions that allow for host and
specifies extensions to [10] that will provide authentication for
currently being proposed
data to be public [10]. If confidentiality of data
it
will not
when they receive their IP configmatioa they can make sure it is correct and that
The
as
it
authenticated and optionally encrypted [19]; this ensures that
is
These protocols may prove
a deliberate design decision by the
IPSec must be used to encrypt
all
data.
to
update messages
be
lighter
[36].
weight than
EETF because they believe
DNS
may also be easier to implement,
IPSec
network traffic. The only reqtoirement is that the TCP/IP stacks of both the host and manager support
not used for security, any additional services added to the system must provide their
own mechanism and
protocol tor authentication.
5 Related
Work
ZEROCONF
5.1
The Intemet Engineering Task
networking
in the
specification
Force's (IETF)
absence of configuration and administiation
of the
protocols, they
have
set
communicate using IP without requiring any
paper
is
the
name
preconfigured
ZEROCONF
a
list
DNS
or
DHCP
servers.
However, when these services are
for
ZEROCONF
prior configuration or the presence
to address resolution problem.
The
In fact, the
present, the
ZEROCONF
ZEROCONF
ZEROCONF
requirements offer temporary and inferior solutions to the
is
currently proposing a protocol to enable
Although they have yet to propose a
[38].
of requirements
Working Groip
protocols [15].
of network
services.
specific implementation or
The
goal
is
to allow hosts to
Of particular relevance
requirements specifically state that there should be no need for
protocols are required to
work
absence of
in the
requirements direct that hosts should use tliem. Thus, the
name
to diis
resolution
problem
until
a complete
name
resolver
is
DNS
servers.
ZEROCONF
located, such as a
DNS server.
ZEROCONF protocols lace two challenges when determining name to address bindings.
and/or hostname on the network. This
is
handled extremely well
in
ZERCONF protocols must not rely on the presence of DHCP server.
IPv4 auto-configuration mechanisms
(useflil
[14].
IPv6 auto configuration
is
modem
The
first is
obtaining a unique address
networks by the use of a
Therefore, the working group
vastly superior as
it
DHCP
server,
recommends using
however,
either
IPv6 or
allows hosts to obtain a link local address
only on a single network) using address auto configuration [14] and a routable address by discovering routei? using Neighbor
26
Discovery
[24].
IPv4 auto configuration
is still in
the research state
by
the IETF, but the
initial
specifications allow a host to only get a
This will prevent the host Irom communicating with any device not directly on the same network as
link-local address [4].
while IFV6 provides nearly all-necessary network parameters such an address, domain name, default router, and
present),
liave a
offers
IPv4 provides only an address. Thus,
new
address, while a host configured
more advantages,
it
is
exjjected
if
will
have a permanent method of contact,
to IPv4 will dominate for some more time [4].
and the lack of fiilly complaint IPv6 routers and hosts on most networks has prevented
Once a
names. The
ZEROCONF
DNS server location (if
a host configured with IPv4 auto configuration leaves a network and
by tPv6
host has obtained an address
on the network,
it still
it
This
is
its
rejoins,
domain name. While IPv6
because IfV6
is
relatively
new
an IP host will also need to
listen for
each case, aU hosts will run a "stub"
name
service that only responds
multicast
when
it
clearly
standard
needs to discover other hosts and resolve domain
such requests and to respond
own name. The IETF currentiy has two ongoing activities in this area:
may
it
fiom gaining widespread acceptance.
ZEROCONF requirements state that hosts should use multicast to resolve names in the absence of a DNS server.
to support this requirement,
Also,
it
In order
when the request corresponds to the host's
DNS [II] and "IFV6 Node Information Queries" [5].
fields a request for its
In
hostname. The stub service does not
provide any name to address resolution for other hosts on the network.
While
several
tiie
naming
service proposed in this paper
key differences. The most obvious one
entirely client devices, with
manager, and provides
configuration, they
ZEROCONF
the
problem
rely
that the agent
that the
on more complete solutions such
no requirements
program attenpts
and managers
that runs the agent
to solve.
the
DNS
broad goals of the Zero Configuration Working Group,
as
DNS
In a
By
network.
it
has
design, the agent
program
is
run on a network
ZEROCONF requirements specify that hosts need no previous
and
DHCP
that these servers
are administrator free and
program on
in the
While the
services for the entire network.
set
into the
ZEROCONF requirements are designed to work with a network composed of
no service providers or managers
Working Group has
paper, both hosts
network
do
DNS
is
may fit
for long-term operation
be self-configuring and
and
scalability.
self-administrating.
However, the
This
is
precisely
network managed by the self-configuring naming service described
may join and
servers
leave the network fieely.
and also meets the
ZEROCONF protocols to obtain an address until the discovery of a manager.
ZEROCONF
Once a manager is
Therefore
it is
requirements.
found, the host
in this
possible to have a
Hosts could use the
is
flee to resolve
any
name on the network using standard DNS calls.
5.2
Non-IP Networks
The requirements
together "just
work"
in the
laid out
by
the
ZEROCONF
Working Group
absence of service providers such as
stress the
DNS and DHCP
[15].
importance of having computers networked
Two protocols in
use today provide
this level
of
27
The AppleTalk
functionality.
will get
of protocols
suite
one a working AppleTalk network
networks can be used
in
is
simple to operate in small networks. Plugging a group of Macs into an Ethernet hub
witliout the
need
homes, schoolrooms, and small
too complicated and costly to administer.
to setup specialized servers like
offices
- environments where
DNS
[14].
As
a consequence, AppleTalk
IP networks have been absent because they are
NetBIOS provides similar fimctionality and ease of use on Microsoft Windows machines.
However, because nearly all computers used today are connected to the Intemet, they also require TCP/IP
is
of the hitemeL Therefore, the benefits of AppleTalk and NetBIOS are overshadowed as application developers
the protocol
to siqjport
two
TCP/IP
protocols:
more
either
AppleTalk or NetBIOS
efficient
In addition, both protocols have issues
name resolution,
alphanumeric
when
thus creating a large
NetBIOS network can be designed such
need
will
This
is
the
that
will
scaling to networks the size
make application development simpler and
of the Intemet AppleTalk
number of unnecessary packets being
relies
on broadcast
A properly configured
sent to each host.
However, NetBIOS
no broadcasts are necessary.
relies
on a
flat
sixteen-chaiacter
name space, which will undoubtedly lead to naming conflicts, particularly when networks are merged
Easing
DNS
The above
Administration
solutions
were
all
over IP or by using different protocols
replacements for a
altogetiier.
network systems, there have been other
repetitive tasks
through automation
fiill
DNS system; eidier by using other methods of name to address mapping
However, because IP networks and
efforts to ease the configuration
[13].
DNS have been integrated into nearly all modem
and administration of DNS.
DNS administration problems by simplifying configuration tasks and eliminating
Researchers in Japan have attempted to tackle
They developed
a program with a graphical interface that reduced the
administrator. Their tool, called nssetiip, automates repetitive tasks such as generation a
keeping the root cache
file
up-to-date,
nssetiip contained a feature that
for configuring the resolver
configure a
to access the local network.
tliis
ZEROCONF Woridng Group and other efforts to make the IP suite of protocols simple to
Allowing developers to concentrate on one protocol for all communications
packets to perform
5.3
and
to access the Intemet,
motivation behind our research as well as the
configure.
be configured, as
to
name
and maintaining the reverse address lookup
checked
if the
name
and adding new hosts
server
was up and
into the database.
server using nssetiip than without
it.
They
table.
work of a
DNS
DNS database file from the machine's host file,
To check
the correctness
of the configuration,
running. In addition, nssetiip provides a graphical user interface
was considerably
faster to
have reduced the configuration time from two hours
to three
The
state tiiey
nssetiip developers
showed
that
it
minutes.
However,
nssetiip
does not truly reach the goal of zero administtation.
default configuration values for
BIND. Every time new DNS
It
simply provides a nice user interface and a good
servers are added, an administrator
set
of
must configure them as well as every
28
DNS server already running on the network so that they are properly integrated into tlie network.
configuration
DNS
when
servers are added.
managers on the network and they
program
5.4
to configure a
name
service
Simply
will configure
is
less
program
themselves accordingly,
ail
is all
that
is
needed to be done;
without user intervention.
than the time using msehip, and does not require a
it
The rime
requires
no
prior
will locate other
it
takes the agent
human administrator.
Commercial Solutions
Tliere are several
Most provide
commercial products available today
solurions that include both
DHCP
corporations that run a large intranet with their
fieely join
and dynamic
own DNS
DHCP
aim
DNS
By
DHCP
in
to simplify the operation
dynamic
including both
will assign the
new
it
commercial solutions with the addition of remote administration.
Check Point Meta
IP 4. 1 and
,
NTS IP Server.
Remote
Commercial
Product
Administration
through
and administrarion of DNS
web browser
DNS
at
and DHCP, they allow hosts
to
host an IP address, and send
Figure
Some
1 1
servers.
These products are aimed
a single software package.
DNS server of the new host so that may be located on the network.
in the
Enterprise 5.0,
that
servers.
and leave network with no manual configuration.
signals to notify
and
starting the agait
The agent program
shows
the integration
DNS
update
between
DNS
popular products include Lucent
QIP
It is
Specifically, the developers
agent
name service runs BIND,
The NTS
IP Server
IP includes
some changes
main
The
server.
other
DNS
protocol to allow for additional functionality.
it
of zone
also supports this feature.
is built
These are
Essentially, "peer
to notify other slaves
irrplementations support the notify message which provides similar functionality [33]. Because the
finm the ground up and
developers to implement two proprietary extensions into the
to Iiave "peer backups".
to the standard
of Meta IP developed a proprietary way for primary and secondary servers
BIND
However, recent
updates.
Meta
important to note that
essentially duplicates
backup" allows
for
is
not based on any previous implementation of
DNS protocol.
of the
server,
The
first
server.
the
DNS extension developed by NTS allows servers
and unlike secondary
more than one primary
BIND. This allowed
Thus
servers, they are
always consistent with the
if one fails, the other is available for updates.
DNS extension allows for zone "co-serving". A zone can be split into a number of pieces each served by a different server.
However, any of the servers
may be queried and may appear to be primary master
other and return the correct answer to the resolver. For larger zones, this
may
for the zone.
They will communicate amongst each
lead to a performance
improvement as the load can be
balanced amongst a number of different servers. However, these extensions are not Internet standards, and only those servers runnuig
the
NTS IP Server will be able to use them
S.4.1
Comparisons
to the agent
name system
In addition the products described above, there exist other similar solutions to ease the administration and operation of
[
1
8][25].
Nearly
all
of these vendor products
support, directory services,
and remote
they do not pertain directly to the
oflFer features that
administration.
name system
DNS
appeal to the corporate environment These features include user profile
While these
features are certainly advantageous in the corporate environment,
In addition, the goal
of these commercial solutions
is
to provide
a central point of
administration for the entire network.
In contrast, the agent
name system
takes a decentralized approach to the administration of the network.
charge of the entire management and most network operations, such as creating a
communication amongst multiple managers. The decentralized approach allows
in the decision
electing a
for the relevant portions
one manager
new
of the network
is
in
master require
to
be involved
making process.
Anotiier key difference between the agent
name
new zone and
No
servers into the systeriL
name system and
those described in this section
is
required
the procedure for introducing
Both the commercial systems and the agent name system require no administrative input when
and leave the network. However, only the agent name system allows servers
reconfiguration
is
to
do the same. In
all
new
hosts join
of the above products, extensive
when a server exits or joins the network.
30
While most of the commercial products provide a method
them provides a pennanent
solution like the agent
name
allow the secondary server to process them. However,
new
system quickly promotes a
commercial solutions must be
to handle
update messages while the primary server
They simply
system.
secondary also
if the
server to primary allowing
all
store the
fails,
updates
the updates will
In the best case, the
administrator a nice graphical user interface for performing the necessary configuration.
necessary
files is
Whatever the case
needed.
be lost
planned so that the master/slave server relationship exists for
carefiilly
is,
name system provides
the agent
down, none of
primary comes back on
updates to be recorded permanently.
configured so that they are aware of the locations of the other servers.
configuration
until the
is
In contrast the agent
line or
name
Networks managed by
all
zones.
the
Servers must be
commercial products provide a human
In the worst case,
a simpler solutioa
manual editing of the
No human
administration
is
when name servers exit or join the network, as the managers will communicate amongst themselves and adapt dynamically to
the changing
network topology.
In
addition,
name system
the agent
solution
works usually
existing
DNS
standards and
implementations. This ensures compatibility and takes advantage of the proven stability of the existing implementations.
Performance Comparison
5.5
The agent program
is
able to autonomously configure a
service configuratioa multiple agent
programs communicate with each other and exchange information, and based on information fi-om
other agents, an agent will generate the named.conff\e and call
services,
and configuration
is
done
DNS process without any human intervention. To accomplish name
in a decentralized inanner.
tiie
This
named command. No human
is
is
required to configure the
name
different fi^om existing configuration solutions in industry
and
research.
Name service solutions
name
server For example, there
system
to
[20]. In
order for the
in industry require
is
much
QIP system
humans
to
manually
install the
server software and configure the server to act as a
planning and modeling involved in the configuration process of the Lucent
to
work
effectively, physical
be determined. This can be a time-consuming process
for a
and
logical relationships
QIP
Enterprise
between network objects and users have
network administrator. With the agent program, the only thing
to
do
is
to
run the agent program.
5.6
Memory Usage of Name System
The two components of the name
memory of the machines used
with 128 Megabytes of RAM.
service, the agent
for testing the
name
system.
program and the
BIND process, took up approximately 2.4% of the total
The machines used
To show the memory usage of the programs,
for testing
the agent
all
had
Intel
Pentium 450
MHz processors
program was run on a machine, and the program
31
lop
was run
to give a
memory and CPU
agent program occupied
1
.7
report.
The
Megabytes of memory and the
Table 4 - Output of top:
Size
usage
Memory
usage of
name
output of the lop program
is
shown
BIND process, nanied, occupied
service
1
.4
in
Table
4.
As shown
Megabytes of memory.
in
Table
4, the
,
[5]
M. Crawford, "IPv6 Node Information Queries", draft-ietf-ipngwg-icmp-name-lookups-07.txt, August
2000.
Droms, "Dynamic Host Configuration Protocol",
[6]
R.
[7]
R. Droms, ed., "Authentication
(work in progress)
[8]
"Dynamic DNS,
for
DHCP
RFC
213
March
1,
1
997.
Messages", draft-ietf-dhc-authentication-I6.txt, January 2001.
Infrastructure essential for today's intranets",
http://www.nts.com/collateral/ddnstcchpaper.pdf
[9]
D. Eastlake,
"DNS
[10]
D. Eastlake,
"Domain Name System
[11]
L Esibov, B. Aboba, and D. Thaler, "Multicast DNS", draft-ietf-dnsext-mdns-OO.txt, November 2000.
[12]
M. Feng, "A Self-Configuring and Self-Administering Name System", http://scannergroup.mit.edu/PDFS/FengM.pdf, Master's Thesis, Massachusetts Institute of Technology, 2001.
[13]
Security Operational Considerations",
RFC
2541, March 1999.
RFC
Security Extensions",
2535, March 1999.
C.S. Giap, Y. Kadobayashi, S. Yamaguchi, "Zero Internet Administration Approach: the case of
Proc. of 12"'
Int.
DNS",
Conf. on Information Net\ivrking (ICOIN), pp. 350-355, 1998.
Guttman, "Zero Configuration Networking", Proceedings ofINET2000, 2000.
[14]
E.
[15]
M.
[16]
K. Homstein,
[17]
P. Huck, "Zero Configuration Name Services for IP Networks", http://scannergroup.mit.edu/PDFS/HuckP.pdf Master's Thesis, Massachusetts Institute ofTechnology, 2001.
[18]
"JOIN DDNS". http://www.ioin.com/ddns.htmI
[19]
S.
[20]
"Lucent QIP Enterprise
Hattig, ed.,
Zeroconf Requirements,
et al,
"DHCP
draft-ietf-zeroconf-reqts-07.txt,
Authentication Via Kerberos V",
March 200
.
(work
in progress)
November 2000.
Kent and R. Atkinson, "Security Architecture for the Internet Protocol",
5.0:
1
RFC
2401, November 1998.
Automating IP Services Management",
http://www.qip.lucent.com/products/qipcnt 6093.pdf
[2
1 ]
"Meta
IP Technical
White Paper", htrp://www. checkpoint.com/products/metaip/whitepaper.pdf
[22]
P.V. Mockapetris,
"Domain Names - Concepts and
[23]
P.V. Mockapetris,
"Domains Names - Implementation and
[24]
T. Narton,
[25]
"Optivity NetID",
Facilities",
RFC
1034,
Specification",
E Nordmark, and W. Simpson, "Neighbor Discovery
December 1998.
November
RFC
1987.
1035 November 1987.
for IP Version 6 (IPv6)",
RFC
2461,
http://www.nortelnetworks.com/products/01/unifiedmanagement/collateral/onetid_brief.pdf
[26]
"Protocol Standard for a
NetBIOS Service on
a
TCP/UDP
Transport: Concepts and
M ethods", RFC
1
00
1
March 1987.
[27]
J.
Reynolds and
J.
Postel,
"Assigned Numbers",
RFC
1700, October 1994.
33
[28]
"Shadow
[29]
G.S. Sidhu, R.F. Andrews, A.B. Oppenheimer, Inside Appletalk, Second Edition, Addison-Wesley
IPserver", http://www.nts.com/collateral/ipserverdatasheet.pdf
.
Publishing
[30]
M
.
Company,
Y
Stapp and
2001. (work
in
.
Inc.,
Rekhter,
1990.
"The
DHCP
Client
FQDN
Option", draft-ietf-dhc-fqdn-option-0
1
March
.txt,
progress)
Doraswamy, and
Document Roadmap",
RFC 241
[31]
R. Thayer, N.
[32]
H. Toivanen, "Secure Zero Configuration", http://citeseer.ni.nec.com/401221.html
[33]
P. Vixie,
"A Mechanism
for
R. Glenn, "IP Security:
Prompt Notification of Zone Changes
1,
November
1998.
.
(DNS NOTIFY)", RFC
1996, August
1996.
"Dynamic Updates
Domain Name System (DNS Update)", RFC 2136,
[34]
P. Vixie, et.al
[35]
P. Vixie, et al,
[36]
B. Wellington, "Secure
[37]
A. Williams, "Securing Zeroconf Networks", draft-williams-zeroconf-securiry-OO.txt,
(work
[38]
.,
"Secret
in
the
Key Transaction
Authentication for
DNS
(TSIG)",
RFC
2845,
May
April
1
997.
2000.
Domain Name System (DNS) Dynamic Update", RFC 3007, November 2000.
November 2000.
in progress)
"Zero Configuration Networking (zeroconO Charter, http://www.ietf org/html. charter.s/zeroconfcharter.html
A Appendix
A.l
Name
Service Information
The agent program stores information about
all
the
name services
messages from the manager and by asking other agents
below.
Server Record Field
in the
in the system.
system The agent obtains
The information
this
information through discover
stored for an individual
name
service
is
shown
agent program uses these four server groups to quickly look up information about the other servers in the network and to exchange
intbrmation with other agent programs.
Server Record List Type
The four groups can have overlapping serva records and are shown below.
Name
IP Address
Server ID
Status Flag
Discover Message
Msgjd
Send/Response Flag
2-2 /Vjif
Date Due
iCC^.
MIT LIBRARIES
DUPL
3 9080 02246 0684
Download