Compaq ProLiant 1000 Compaq Backup and Recovery for Microsoft SQL Server 6.X - Page 38

Compaq Backup and Recovery for Microsoft SQL Server 6.x, Configuring Compaq RAID, Technology

Get Compaq ProLiant 1000 PDF manuals and user guides View all Compaq ProLiant 1000 manuals
Add to My Manuals
Save this manual to your list of manuals

Page 38 highlights

Compaq Backup and Recovery for Microsoft SQL Server 6.x Chart 5 - Striped Dumps to Disk Arrays Page 38 As expected, performance increases as the number of storage devices (controllers, SCSIports, spindles) used is increased. The performance trend when dumping to RAID-0 arrays however, differs from that seen with RAID-5 arrays. For the RAID-0 arrays, when the number of storage devices is doubled from the first test to the second - throughput jumps about 50% (from about 32 to 48 GB/hr). After that only a marginal increase is seen, because at this point the limit of the source (read) controller to deliver data is already reached. With a throughput of nearly 50 GB/hr on the third test however, we have reached the maximum throughput predicted in the previous section (allowing for a small additional overhead for write processing)61. Performance for the RAID-5 arrays at first lags significantly behind that of the RAID-0, but scales quickly as storage devices are added. The performance doubles as the number of storage devices are doubled from the first test to the second (from 16 to 32 GB/hr). This is because the throughput limitation of the source controller has not yet been reached - at this point our bottleneck is at the Smart2 controllers used to write the data to; RAID-5 is very write-intensive and tends to saturate the data channels on the controller when large (64KB) writes requests are rapidly being issued62. From the second to the third tests, the number of controllers has been doubled again, so that the performance 61 The throughput for the 3rd RAID-0 test may have yielded a higher number if SQL Server had been reading the database from two Smart-2 controllers combined into a Windows NT stripe set, as with the 3rd null device test in the previous section. 62 Writes to a RAID level 5 disk array may involve up to 4x additional I/O. When transferring data to a RAID-5 volume, 4 physical I/O's (2 reads and 2 writes) must normally occur for every logical write request so that the parity information may be re-created. If purely sequential writes occur to the array however, the Smart-2 controller can sometimes recognize the contiguous data pattern and thus perform fewer additional I/O's for the parity (i.e: parity can be generated for each "stripe" across all drives instead of for each write to any one drive). For a discussion on the performance implications of RAID technology, see the Configuring Compaq RAID Technology for Database Servers white paper.

  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
  • 47
  • 48
  • 49
  • 50
  • 51
  • 52
  • 53
  • 54
  • 55
  • 56
  • 57
  • 58
  • 59
  • 60
  • 61
  • 62
  • 63
  • 64
  • 65
  • 66
  • 67
  • 68
  • 69
  • 70
  • 71
  • 72
  • 73
  • 74
  • 75
  • 76
  • 77
  • 78
  • 79
  • 80
  • 81
Compaq Backup and Recovery for Microsoft SQL Server 6.x
Page
38
Chart 5 - Striped Dumps to Disk Arrays
As expected, performance increases as the number of storage devices (controllers, SCSIports, spindles)
used is increased.
The performance trend when dumping to RAID-0 arrays however, differs from that
seen with RAID-5 arrays.
For the RAID-0 arrays, when the number of storage devices is doubled from
the first test to the second - throughput jumps about 50% (from about 32 to 48 GB/hr).
After that only
a marginal increase is seen, because at this point the
limit of the source (read) controller to deliver
data
is already reached.
With a throughput of nearly 50 GB/hr on the third test however, we have
reached the maximum throughput predicted in the previous section (allowing for a small additional
overhead for write processing)
61
.
Performance for the RAID-5 arrays at first lags significantly behind that of the RAID-0, but scales
quickly as storage devices are added.
The performance doubles as the number of storage devices are
doubled from the first test to the second (from 16 to 32 GB/hr).
This is because the throughput
limitation of the source controller has not yet been reached - at this point our bottleneck is at the Smart-
2 controllers used to write the data to; RAID-5 is very write-intensive and tends to saturate the data
channels on the controller when large (64KB) writes requests are rapidly being issued
62
.
From the
second to the third tests, the number of controllers has been doubled again, so that the performance
61
The throughput for the 3rd RAID-0 test may have yielded a higher number if SQL Server had been reading
the database from two Smart-2 controllers combined into a Windows NT stripe set, as with the 3rd null device
test in the previous section.
62
Writes to a RAID level 5 disk array may involve up to 4x additional I/O.
When transferring data to a RAID-5
volume, 4 physical I/O’s (2 reads and 2 writes) must normally occur for every logical write request so that the
parity information may be re-created.
If purely sequential writes occur to the array however, the Smart-2
controller can sometimes recognize the contiguous data pattern and thus perform fewer additional I/O’s for the
parity (i.e: parity can be generated for each “stripe” across all drives instead of for each write to any one drive).
For a discussion on the performance implications of RAID technology, see the
Configuring Compaq RAID
Technology for Database Servers
white paper.