BACKUP/MASTER:
Immediate Relief with
Disk Backup
Presented by
W. Curtis Preston
VP, Service Development
GlassHouse Technologies, Inc
Tape backups are taking too long
 High-speed tape drives in a library are the standard,
but the cost of these units causes many people to
cut corners elsewhere
 The nature of tape drives also creates difficulty
when creating offsite tapes
 Many people aren’t utilizing the tape drives properly
and are not getting all their backups done
 Also, many are not creating offsite copies
 Stand-alone tape drives must be swapped
Tape drives: The advantages
 High-speed, low cost
 Good archival solution. Allows multiple
copies without significant cost.
 Lots of new tape drives on the market:
•
•
•
•
9940B (30/70 MB/s)
AIT-3 (15/30 MB/s)
LTO (15/30 MB/s)
Super DLT (11/22 MB/s)
Tape: The challenges
 Tapes are now too fast!
• Must use multiplexing to stream them during
network backups
• Must use higher multiplexing values than ever
before, hurting restore performance more
 Tape-to-tape copying takes time, and multiplexing
increases that time – especially if you de-multiplex
 Must perform regular full backups to reduce number
of tapes required for restore
 Incremental backups do not supply enough data to
stream a tape drive
Tape: The challenges (2)
 Cannot write to single tape drive from two
shared servers simultaneously
 Single tape can cause large restore to fail
 You never know if a tape is good until you
really need it
Still not making offsite copies
 Assuming copy is same speed as backup,
must buy at least twice as many drives to
perform copies in one day
 If copy is not same speed, must accept
longer copy window or buy more tape drives
 Additional drives cost a lot of money
 Result: Many people still not making offsite
copies
Solution: New backup media
 Really inexpensive disk arrays
•
•
IDE/ATA-based
•
JBOD and RAID configurations (Use their
RAID controller or a software volume
manager.)
•
As low as $5,000 per TB for off-shelf units,
$2,000 for build-your-own units!
Addressable via Fibre Channel, SCSI, Firewire,
NFS, or CIFS
What to do with them?
 Buy enough disk for two full backups and
many, many incremental backups
 Connect array to clients or backup
servers via Fibre Channel & SANs, or GbE
& NFS/CIFS
Backup
Client
Backup
Server
NFS/CIFS/SAN
ATA Disk
Array
Copy or second
backup
Tape
What to do with them? (2)
 Back up to disk first using your backup
software of choice
 Duplicate disk backups to tape
 Except in disaster, restores come from disk
 Maybe place (another?) disk unit offsite
and replicate to it
Backup
Client
Backup
Server
NFS/CIFS/SAN
ATA Disk
Array
Copy or second
backup
Tape
Offsite
Offsite
ATA Disk
Array
ATA Disk
Array
What to do with them? (3)
 Most backup products do things that are
not necessary when backing up to disk
•
•
•
Occasional full backups
Backing up redundant files
Incremental backups of entire files
 New products designed to back up to disk
•
•
Forever incremental w/o performance hit
Some even eliminate redundant blocks across hosts
What to do with them? (4)
 Replicate many
clients to a central
array, back up that
array using backup
software, and
duplicate to tape for
offsite copies
 Allows you to use
replication without
the cost of traditional
RAID arrays
Backup
Client
Backup
Client
Backup
Client
Replication
Server
Tape
ATA Array
What to do with them? (5)
 Could also use software-based RAID to
create additional mirror, and split mirror for
backups
 Gives you BCV functionality for ¼ the price!
 Back up large databases with no I/O
overhead on server!
Why would you do that?
 Don’t require constant stream
 No need to multiplex on most disk devices
 Depending on implementation, multiplexed
backups may still be faster on disk
 If you did multiplex your disk backups, you
could easily de-multiplex the tape copies
with no performance penalty
 NFS/CIFS devices can be used
simultaneously by many clients, without
needing to stream each device
Why would you do that? (2)
 Incremental backups with little data will not
hurt performance of other backups
 Protected via monitored RAID -- the loss of a
single disk would be monitored and repaired,
while the RAID group continued to protect
the data
 Disk-to-tape copies are easier than tape-totape copies
 Could perform infrequent full backups
without increasing the chance of failure
 Full backups can be performed less often,
saving networks and CPU utilization
Why not back up everything to disk?
 Archiving purpose of backups requires
older backups to be available
 Tapes still much cheaper, allowing for
multiple, stable copies to be put on “the
shelf” onsite or offsite
 Tapes not susceptible to filesystem
corruption
Issues…
 Staging process needs automation
•
•
Need to automatically move data from disk to tape without
removing from disk
Should allow you to leave backups on disk ALAP, and
automate moving data to tape when necessary (policybased, not just retention-based.)
 Increase ease of recovery
•
Need to be able to import disk images
 Creation of a “Synthetic Full” would be very
nice
 Backup twinning should be able to go to disk
and tape
In Short
 Doing backups to inexpensive disk first
allows for:
•
•
•
•
Faster, easier backups – especially incremental
backups
Easier creation of offsite tapes
Easier restores both on- and offsite
Many other features
 A directory of ATA Fibre & SCSI
addressable arrays is available at:
http://www.storagemountain.com
 Questions to cpreston@glasshouse.com