HP ProLiant 1500 Compaq Backup and Recovery for Microsoft SQL Server 6.X - Page 39

available through the Compaq Resource Kit for Windows NT.

Page 39 highlights

Compaq Backup and Recovery for Microsoft SQL Server 6.x Page 39 goes up almost 50% to 47 GB/hr which is close to what it was with RAID-0. The average CPU usage seen during the third set of tests (for both types of arrays) was around 60%. In this section data is sent to multiple 15/30-GB DLT63 tape drives, using the tape drive striping functionality of the SQL Server dump database process. For these types of dumps to multiple tape drives, the following considerations were taken into account: As before, the database was contained on a single RAID-5 disk array, using a single Smart-2/P controller (both SCSI-ports) which resided on the primary PCI bus of the Proliant 5000. All of the DLT drives resided in Compaq DLT Tape Array cabinets (4 drives per cabinet). Each cabinet was interfaced to 2 controllers (the embedded Fast-Wide SCSI-2/P controller or additional Wide-Ultra SCSI-2/P cards). Up to 12 tape drives and 6 controllers were used in the testing, with no more than 2 drives attached to any one controller.64 All tape drives are used as part of a single tape device 'stripe set'. Since up to 12 tape drives were used, the backup_threads parameter was set accordingly, as the SQL Server dump process uses one thread per device. Compression was enabled on the tape drives, providing an approximately 15% increase in throughput vs. non-compressed. A database containing data that lends itself differently to compression could yield throughput somewhat higher or lower than that displayed below. Using the above guidelines, performance was measured using backup sets of 1 to 12 DLT drives. Since we expect to get a high throughput rate from 12 DLT's, we monitor the hardware channels on the system (PCI bus65 and CPU usage) to ensure that a performance bottleneck is not encountered elsewhere. The results are shown in the following chart: 63 The results in this section can be applied to 10/20-GB DLT tape drives as well, since the performance of these drives is the same as that of the 15/30 drives. 64 The 8-bit, asynchronous mode SCSI transfers done by the DLT will allow us to practically use no more than about 3 MB/sec of the SCSI bus bandwidth, and so throughput limitations are encountered beyond 2 drives on a controller. With 12 drives however, the throughput maintained to each individual drive falls below even 1 MB/sec, so that placing 3 drives per controller can be considered. 65 PCI bus usage was monitored using special Performance Monitor counters developed by Compaq, which are available through the Compaq Resource Kit for Windows NT. The PCI % utilization shown in the chart was taken as an average of the load observed between the 2 peer PCI buses. 1997 Compaq Computer Corporation, All Rights Reserved Doc No 444A/0797

  • 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
39
1997 Compaq Computer Corporation, All Rights Reserved
Doc No 444A/0797
goes up almost 50% to 47 GB/hr which is close to what it was with RAID-0.
The average CPU usage
seen during the third set of tests (for both types of arrays) was around 60%.
In this section data is sent to multiple 15/30-GB DLT
63
tape drives, using the tape drive striping
functionality of the SQL Server
dump database
process.
For these types of dumps to multiple tape
drives, the following considerations were taken into account:
As before, the database was contained on a single RAID-5 disk array, using a single Smart-2/P
controller (both SCSI-ports) which resided on the primary PCI bus of the Proliant 5000.
All of the DLT drives resided in Compaq DLT Tape Array cabinets (4 drives per cabinet).
Each
cabinet was interfaced to 2 controllers (the embedded Fast-Wide SCSI-2/P controller or additional
Wide-Ultra SCSI-2/P cards).
Up to 12 tape drives and 6 controllers were used in the testing, with
no more than 2 drives attached to any one controller.
64
All tape drives are used as part of a single tape device ‘stripe set’.
Since up to
12
tape drives were
used, the
backup_threads
parameter was set accordingly, as the SQL Server dump process uses
one thread per device.
Compression was enabled on the tape drives, providing an approximately 15% increase in
throughput vs. non-compressed.
A database containing data that lends itself differently to
compression could yield throughput somewhat higher or lower than that displayed below.
Using the above guidelines, performance was measured using backup sets of 1 to 12 DLT drives.
Since
we expect to get a high throughput rate from 12 DLT’s, we monitor the hardware channels on the
system (PCI bus
65
and CPU usage) to ensure that a performance bottleneck is not encountered
elsewhere.
The results are shown in the following chart:
63
The results in this section can be applied to 10/20-GB DLT tape drives as well, since the performance of these
drives is the same as that of the 15/30 drives.
64
The 8-bit, asynchronous mode SCSI transfers done by the DLT will allow us to practically use no more than
about 3 MB/sec of the SCSI bus bandwidth, and so throughput limitations are encountered beyond 2 drives on a
controller.
With 12 drives however, the throughput maintained to each individual drive falls below even 1
MB/sec, so that placing 3 drives per controller can be considered.
65
PCI bus usage was monitored using special Performance Monitor counters developed by Compaq, which are
available through the Compaq Resource Kit for Windows NT.
The PCI % utilization shown in the chart was
taken as an average of the load observed between the 2 peer PCI buses.