Compaq ProLiant 1000 Compaq Backup and Recovery for Microsoft SQL Server 6.X - Page 37

Smart-2/P Array controllers.

Page 37 highlights

Compaq Backup and Recovery for Microsoft SQL Server 6.x Page 37 array could conceivably be used for storage of the data without an accompanying tape device57. In this case a fault tolerant array would be highly recommended, preferably on a separate controller from the one containing the actual database volume. In this section backup data is sent to SQL Server disk dump devices, which are created on additional Smart-2/P Array controllers. For these types of dumps to local disk array controllers, the following modifications were made to the system configuration: External storage cabinets housing 5 Wide SCSI drives each were attached to each of the Smart-2/P controllers used to store the backup data. Smart-2 controllers used to hold the disk dump devices were placed on the secondary PCI bus of the Proliant 5000, while the controllers holding the database itself were placed on the primary PCI bus. In this way, system throughput was optimized because the load was split evenly across the peer PCI channels58 (all data read into system memory from peripherals on PCI bus 1, all data written out from memory to peripherals on PCI bus 2).59 A single disk array was configured on each SCSI port of each controller. A single disk dump device was created under SQL Server for each array, so that only one stream of writes will occur to any one array. This way, sequential writes are enforced on each array even though multiple dump devices are used. You can create more than one dump device file per array, but then the writes to this array will become randomized and performance may suffer (due to disk head contention). Backup 'stripe sets' were created with the dump devices. The SQL Server dump process uses one 'thread' per dump device, so that either 2 or 4 threads are being used depending on the number of devices. (In Chart-5, this works out to #threads = #SCSIports). Destination (dump) disk arrays were built using RAID-5 fault tolerance, then again with RAID-0 for better performance. A minimum 2:1 ratio is maintained at all times between the number of SCSIports/disks used for writes vs. for reads (i.e: there are always at least twice the number used for the dump devices as for the data array). This is to prevent any 'funnel' effect on the writes, since the Smart-2 can perform sequential reads faster than it can complete large, sequential writes. The on-board cache of the Smart-2 controller(s) used for the dump devices was dedicated for write operations60. Using the above guidelines, performance was measured for three different system layouts - using both the RAID-0 and RAID-5 array dump devices (6 tests total). The results are shown in the below chart: 57 The feasibility of using disk volumes for incremental backups is enhanced by SQL 6.x's ability to append data to disk dump devices. 58 Exception: for the last set of tests in Chart-10, 1 read controller and 2 write controllers were configured on each of the PCI buses (3 controllers per bus). 59 If the peripheral controllers are placed so that all data transfers occur on a single PCI bus, throughput may be limited by the PCI bus bandwidth. 60 The on-board controller cache was manually configured for 100% write caching (vs. read-ahead) using the Array Configuration Utility. However, this adjustment would be made dynamically by the controller logic if no read activity was detected on the volume. 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
37
1997 Compaq Computer Corporation, All Rights Reserved
Doc No 444A/0797
array could conceivably be used for storage of the data without an accompanying tape device
57
.
In this
case a fault tolerant array would be highly recommended, preferably on a separate controller from the
one containing the actual database volume.
In this section backup data is sent to SQL Server disk dump devices, which are created on additional
Smart-2/P Array controllers.
For these types of dumps to local disk array controllers, the following
modifications were made to the system configuration:
External storage cabinets housing 5 Wide SCSI drives each were attached to each of the Smart-2/P
controllers used to store the backup data.
Smart-2 controllers used to hold the disk dump devices
were placed on the secondary PCI bus of the Proliant 5000, while the controllers holding the
database itself were placed on the primary PCI bus.
In this way, system throughput was optimized
because the load was split evenly across the peer PCI channels
58
(all data read into system memory
from peripherals on PCI bus 1, all data written out from memory to peripherals on PCI bus 2).
59
A single disk array was configured on each SCSI port of each controller.
A single disk dump
device was created under SQL Server for each array, so that only one stream of writes will occur to
any one array.
This way, sequential writes are enforced on each array even though multiple dump
devices are used.
You can create more than one dump device file per array, but then the writes to
this array will become randomized and performance may suffer (due to disk head contention).
Backup ‘stripe sets’ were created with the dump devices.
The SQL Server dump process uses one
‘thread’ per dump device, so that either 2 or 4 threads are being used depending on the number of
devices.
(In Chart-5, this works out to #threads = #SCSIports).
Destination (dump) disk arrays were built using RAID-5 fault tolerance, then again with RAID-0
for better performance.
A minimum 2:1 ratio is maintained at all times between the number of SCSIports/disks used for
writes vs. for reads (i.e: there are always at least twice the number used for the dump devices as for
the data array).
This is to prevent any ‘funnel’ effect on the writes, since the Smart-2 can perform
sequential reads faster than it can complete large, sequential writes.
The on-board cache of the Smart-2 controller(s) used for the dump devices was dedicated for write
operations
60
.
Using the above guidelines, performance was measured for three different system layouts - using both
the RAID-0 and RAID-5 array dump devices (6 tests total).
The results are shown in the below chart:
57
The feasibility of using disk volumes for incremental backups is enhanced by SQL 6.x’s ability to append data
to disk dump devices.
58
Exception: for the last set of tests in Chart-10, 1 read controller and 2 write controllers were configured on
each of the PCI buses (3 controllers per bus).
59
If the peripheral controllers are placed so that all data transfers occur on a single PCI bus, throughput may be
limited by the PCI bus bandwidth.
60
The on-board controller cache was manually configured for 100% write caching (vs. read-ahead) using the
Array Configuration Utility.
However, this adjustment would be made dynamically by the controller logic if no
read activity was detected on the volume.