Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
Virtualization Technology for Software Rejuvenation
and High Performance Computing
V.DharmaPrakash, Dr.C.Kavitha
Lecturer – Dept of MCA, Professor – Dept of MCA
K.S.Rangasamy College of Technology, Tiruchengode
Abstract
The phenomenon that the state of software degrades with time is known as
software aging. The primary method to fight aging is software rejuvenation. This paper
presents new ways of effective software rejuvenation using virtualization for addressing
software aging and also for High Performance Computing. This new approach is meant
to be the less disruptive as possible for the running service and to get a zero downtime in
most of the cases. Virtualization is a common strategy for improving the utilization of
existing computing resources, particularly within data centers. However, its use for high
performance computing (HPC) applications is currently limited despite its potential for
both improving resource utilization as well as providing resource guarantees to its users.
Keywords: Software aging, Software rejuvenation, Virtualization, High Performance
Computing
degradation
1. Introduction
As
business
the
exhaustion
system
increasingly dependent on information
corruption
and
and computing technology, continuous
accumulation. Aging has not only been
availability is a universal concern. It has
observed in software used on a mass
now been well-established that failures
scale but also in specialized software
of computer systems are more often due
used in high availability and safety
to software faults than due to hardware
critical applications. The most natural
faults.
wide-spread
procedure to counteract such software
phenomenon of “software aging”, one in
aging is to apply the well-known
which the state of a software system
technique of software rejuvenation. It
gradually
degrades
involves
eventually
leads
the
with
and
numerical
occasionally
stopping
data
error
the
performance
software application, cleaning its internal
degradation, transient failures or even
state and/or its environment, and then
crushes
been
restarting it. By removing the accrued
reported. The primary causes of this
error conditions and freeing up or
of
to
time
resources,
of
operating
Recently,
becomes
are
applications,
has
Copy Right @CSE/IT/ECE/MCA-LVEC-2009
Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
decrementing operating system (OS)
environments and it is emerging as a
resources, this technique proactively
technique
prevents
performance.
unexpected
future
system
outages.
to
increase
system
2. Proposed System:
Virtualization technology enables
In the proposed system, it can be able
a single computer to run multiple
to optimize the rejuvenation operation
operating
without
systems
Virtualization
simultaneously.
any
additional
find
hardware. The proposed system is
important applications over a wide range
totally supported by software and can
of areas such as server consolidation,
be easily deployed. In the proposed
secure computing platforms, supporting
system (refer figure.1), on top of
multiple OS, kernel debugging and
virtualization middleware layer, there
development, system migration, etc,
are
resulting
usage.
application server. A software load-
Virtualization allows abstracting away
balancer (VM-LB) will be run on
the real hardware configuration of a
VM1 that will be responsible for the
system. One method of virtualizing the
detection of software aging or some
hardware
computer
potential anomaly. We use the active
involves using a layer of software, called
VM(primary VM) to run the main
the Virtual Machine Monitor (VMM), to
application server and a standby VM
provide the illusion of real hardware for
where we instantiate a replica of the
multiple virtual machines (VMs). Inside
application server that works as a hot-
each VM, the operating system (often
standby. In hot-standby configuration,
called the guest OS ) and applications
component state is replicated to the
run on the VM’s own virtual resources
standby on any change, i.e. the
such as virtual CPU, virtual network
standby component is always up-to-
card, virtual RAM, and virtual disks. A
date. In case of a failure, the standby
VMM can be hosted directly on the
component
computer hardware or within a host
component and continues to operate
operating system. Virtual machines offer
based on the current state. The hot-
a degree of flexibility that is not possible
standby
to
continuous availability without any
obtain
technologies
requiring
in
widespread
resources
on
of
physical
a
machines.
three
virtual
machines
replaces
the
configuration
service.
failed
offers
Virtualization has proved as a successful
interruption
tool for management of complex IT-
Detector and Anomaly-Detector will
Copy Right @CSE/IT/ECE/MCA-LVEC-2009
of
per
Aging-
Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
be used for the detection of software
implementation
aging or some potential anomaly.
technology for HPC.
When
software
aging
or
some
of
virtualization
3.1 For software rejuvenation action
potential anomaly happens VM-LB
The assumptions used in the modeling
will trigger a rejuvenation operation.
are as follows:
In VM1, other necessary software
 Failure rate and repair rate of the
modules will be installed. In order not
VM are identical at all states.
to lose any in-flight request or session
 Unstable rate, the speed of
data at the time of rejuvenation, first,
escaping the healthy condition of
we start the standby server; all the
VM is identical at all states.
are
 Rejuvenation rate, the frequency
migrated from the active server to
of rejuvenation is identical at all
standby server. When all the ongoing
states.
new
requests
and
sessions
requests are finished in primary
 The probability of going from
server, then the primary VM will be
normal state to failure state is
rejuvenated.
negligible compared to other
probabilities.
 During the rejuvenation process,
the system
can provide the
continuous service except non
virtualized scenario.
3.2 For High Performance Computing
Most
virtualization strategies fall into one of
four major categories:
3.2.1.
Full
Virtualization:
Also
sometimes called hardware emulation. In
this case an unmodified operating system
is run using a hypervisor to trap and
Figure 1
3. System Model : In this section, first
safely
determining
virtualization
instructions on-the-fly. Because trapping
technology can improve the software
the privileged instructions can lead to
rejuvenation
significant performance penalties, novel
how
action
and
then
translate/
execute
privileged
strategies are used to aggregate multiple
Copy Right @CSE/IT/ECE/MCA-LVEC-2009
Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
instructions and translate them together.
securely isolate multiple instances of an
Other enhancements, such as binary
operating system within a single host
translation,
improve
machine. The guest operating system
performance by reducing the need to
instances are often referred to as virtual
translate these instructions in the future.
private servers (VPS). The advantage to
3.2.2. Paravirtualization:
operating system-level virtualization lies
can
virtualization,
further
Like full
paravirtualization
also
mainly in performance. No hypervisor /
uses a hypervisor, and also uses the term
instruction trapping is necessary. This
virtual machine to refer to its virtualized
typically results in system performance
operating systems. However, unlike full
of near-native speeds. The primary
virtualization, paravirtualization requires
disadvantage is that all VPS instances
changes to the virtualized operating
share a single kernel. Thus, if the kernel
system.
to
crashes or is compromised, all VPS
coordinate with the hypervisor, reducing
instances are compromised. However,
the use of the privileged instructions that
the advantage to having a single kernel
are typically responsible for the major
instance is that fewer resources are
performance
consumed due to the operating system
This
allows
the
penalties
VM
in
full
virtualization. The advantage is that
overhead of multiple kernels.
paravirtualized virtual machines typically
3.2.4. Native Virtualization: Native
outperform
virtual
virtualization leverages hardware support
machines. The disadvantage, however, is
for virtualization within a processor itself
the need to modify the paravirtualized
to aid in the virtualization effort. It
virtual machine / operating system to be
allows multiple unmodified operating
hypervisor-aware. This has implications
systems to run alongside one another,
for operating systems without available
provided that all operating systems are
source code.
capable of running on the host processor
fully
virtualized
System-level
directly. That is, native virtualization
Virtualization: The most intrusive form
does not emulate a processor. This is
of virtualization is operating system level
unlike the full virtualization technique
virtualization.
para
where it is possible to run an operating
virtualization and full virtualization,
system on a fictional processor, though
operating
virtualization
typically with poor performance. In x86,
does not rely on a hypervisor. Instead,
64 series processors, both Intel and
the operating system is modified to
AMD support virtualization through the
3.2.3.
Operating
Unlike
system-level
both
Copy Right @CSE/IT/ECE/MCA-LVEC-2009
Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
Intel-VT and AMD-V virtualization
extensions. x86, 64 processors with
Since the type-I comparable to that of
virtualization
relatively
native execution. In contrast, type-II
recent, but are fast becoming widespread.
virtualization incurs additional overhead
Virtualization creates an abstraction of
due to the layering of the VMM on top
the hardware and executes one or several
of the host OS when servicing resource
virtual machines (VMs) on top of this
requests from VMs. The type-II layering
virtualized hardware; in some instances,
makes its approach more suitable for the
the virtual machine may also directly
development
access the hardware for performance
performance
purposes. Virtualization solutions are
exchange for greater diagnostic and
today based on the concept of a Virtual
development capabilities. Today, several
Machine Monitor (VMM), also called a
system-level virtualization solutions are
hypervisor. The VMM is responsible for
available, for instance Xen (type-I),
the hardware virtualization and execution
QEMU
of VMs on top of the virtualized
workstation & server (type-II). However,
hardware. The VMM is typically a small
these solutions are not suitable for HPC
operating system that does not include
because they were not designed to meet
hardware drivers. To access physical
the specialized needs of HPC. For
resources, the VMM is typically coupled
instance, Xen has become a rather
with a standard operating system, which
massive
provides device/hardware access. There
unneeded features for HPC, e.g., a
are
employed,
network communication bus; QEMU and
formalized by Goldberg in the 1970’s
VMWare do not support direct access to
as: (i) type-I virtualization where the
high-performance
VMM and VM run directly on the
Virtualization solution for HPC requires
physical
only a small set of system services, such
two
support
are
approaches
hardware, and (ii) type-II
phase,
may
where
be
(type-II),
micro-kernel
migration,
some
reduced
or
that
network
in
VMWare
includes
solutions.
virtualization where the VMM and VM
as
suspend/resume,
and
run on a host operating system.
checkpoint / restart of VMs. In addition,
the solution should afford developers
Host OS VM VM
M
M
VMM
Hardware
Type I
VM VM
M
M
VMM
Host OS
Hardware
efficient access to resources, a VM
adapted
scheduling policy,
and
be
lightweight in order to: (a) minimize the
system footprint, and (b) guarantee
Type@CSE/IT/ECE/MCA-LVEC-2009
II
Copy Right
Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
performance prediction and isolation for
available for the virtual server while
running VMs.
other can not do this.
Based on usage criteria, we can
vi)Security:-
classify the problems as follows :
techniques increases security by allowing
i) Multi OS:-
Some virtualization
system administrators to cleanly separate
technology only support a type of OS
services on different virtual servers,
(Linux, Windows, etc.) while others are
some
of
them
more generic and can run Linux on
protections
with
Windows, Windows on Linux, etc.
additional security model that can make
MultiOS virtualization systems include
a virtual server more robust than a real
VMware and Xen.
one.
ii) Kernel Development/Debugging :-
Hypervisor for High Performance
Computing
Some users need to develop the kernel.
This criteria will define if,yes or no,
those tasks can be achieved with the
iii) OS Installation Process :- Some
users need to reproduce the complete
installation of a system (Install CD,
Network Boot, Hard disk partitioning,
consumption:-
This
criteria will define how much resources a
virtual computer need to use in order to
be
fully
virtualization
functional.
For
technique,
offers
additional
rules/roles
and
The core of a virtualization solution for
HPC is a HPC Hypervisor. Current
server consolidation market. Because of
this context and because of the lack of
hardware
current
support
VMMs
footprint.
etc.).
Resources
virtualization
VMM has been initially designed for the
chosen virtualization technique.
iv)
All
each
the
approximative resource consumption of a
fully functional virtual server has been
v) Dynamical allocation of resources:Some users needs to dynamically change
the resources used by a virtual computer.
Some virtualization programs allow the
user to change live the resources
have
virtualization,
huge
important
system
system
footprint is created by two different
factors: (i) the VMM is always coupled
to a HostOS which is a full operating
system, (ii) for virtualization purpose
the VMM store a large dataset
in
memory.
VMM
Memory
system-level
estimated.
This
for
Footprint
virtualization
Typical
solutions
typically have a “view” of the virtual
machine memory space. This view is
used to translate the memory access from
the virtual machine to the physical
memory. This memory footprint has a
direct impact on modern execution
Copy Right @CSE/IT/ECE/MCA-LVEC-2009
Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
platform which are most of the time
the VMM itself. The system footprint is
composed of multicore processors with
actually
shared cache. Typically if the core non-
HostOS. In order to achieve high
shared cache is smaller than the VMM
performance computing, the HostOS
memory footprint, and if the application
footprint has to be minimized in order to
is memory constrained, cache misses and
limit the interference with the application
flushes generated by the VMM execution
execution. This effort is similar to the
will
the
standard OS tuning effort for HPC
application execution (generating a lot of
platforms, which includes eviction of
cache misses and flushes). In order to
unnecessary daemons and services. Each
address this issue, two approaches are
OS has its own advantages and targeted
possible:
scientific applications. Flexibility in OS
(i) assign the VMM to a specific core
deployment
and decrease its memory footprint to
applications need to be able to run on the
make it fit in the none-shared memory
OS
associated to the core where the VMM is
Virtualization technology is capable of
running, (ii) include some hardware
providing
support to store part of the VM’s
application
“memory map” directly on the hardware.
deployment, and customized system
System
environments
interfere
directly
Footprint
of
with
the
Tuple
mostly
composed
is
they
needed
were
by
as
these
designed
testbeds
for
for.
OS
and
development
for
the
and
applications
by
VMM/HostOS : The VMM needs to be
allowing the interchanging of the OS and
coupled to an HostOS which is needed
its
to host the hardware drivers. Everytime
component on demand.
an application running inside a virtual
4. Conclusion
Runtime
Environment
(RTE)
machine wants to access the hardware
We conclude that virtualization can
(e.g. for network communications), it has
be helpful for software rejuvenation and
to go be default through the HostOS .
fail-over in the occurrence of application
The system footprint of a system-level
failures
virtualization solution is therefore not
virtualization
only composed of the system footprint of
performance computing.
Thandar Thein, Sung-Do Chi, and Jong Sou
Park
“Improving
Fault
Tolerance
by
Virtualization and Software Rejuvenation”- IEEE
Transactions on
software
aging
and
solution
for
high
Conference on Modelling & Simulation pp 855 –
References
1.
and
Second
Asia
International
860, 2008
2. Geoffroy Vall´ee , Thomas Naughton ,
Christian Engelmann , Hong Ong and Stephen L.
Scott “System-Level Virtualization for High
Performance Computing” - IEEE Transactions
Copy Right @CSE/IT/ECE/MCA-LVEC-2009
Proceedings of the International Conference , “Computational Systems and Communication
Technology” Jan.,9,2009 - by Lord Venkateshwaraa Engineering College,
Kanchipuram Dt.PIN-631 605,INDIA
on 16th Euromicro Conference on Parallel,
Distributed and Network-Based Processing pp
636 – 643, 2008.
3. J. P. Walters, Vipin Chaudhary, and Minsuk
Cha , Salvatore Guercio Jr. and Steve Gallo “A
Comparison of Virtualization Technologies for
HPC
”
-
International
IEEE
Transactions
Conference
on
on
22nd
Advanced
Information Networking and Applications pp 861
– 868, 2008.
4. Benoˆıt des Ligneris “Virtualization of Linux
based computers ” Proceedings of the 19th
International Symposium on High Performance
Computing Systems and Applications .
5. M. Grottke, Lei Li, K. Vaidyanathan and K.S.
Trivedi, “Analysis of SoftwareAging in a Web
Server”, IEEE Transactions on Reliability, Vol. 5,
No. 3, September 2006.
6. Virtual Machines: Online Available:
http://www.vmware.com
Copy Right @CSE/IT/ECE/MCA-LVEC-2009