Compaq ProSignia 500 Compaq Backup and Recovery for Microsoft SQL Server 6.X - Page 37
Smart-2/P Array controllers.
View all Compaq ProSignia 500 manuals
Add to My Manuals
Save this manual to your list of manuals |
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