User`s guide

Ultra 320 SCSI, or U320 supports up to 320MB/s throughput per channel. The HP SmartArray 6404 can

support multiple U320 channels (four in the case of the SA6404) and on-board, battery-backed-up cache.

The cache provides for increased read and write performance because it allows the controller to batch

requests to the drives.

Serial Attached SCSI (SAS) uses special 2.5” drives, and has a higher data rate than the U320, up to

600MB/s for newer controllers. SAS controllers typically have more ports than the channels in U320

controllers. Ports and channels are similar because they provide parallel data paths through the

controller. For example, an HP P600 provides 8 ports and each port is capable of 300MB/s.

For U320, the interface capacity is often the bottleneck. With SAS, the drives themselves can be the

bottlenect. SAS seems to provide better performance, flexibility, and scalability than U320 for VCM

installations. In addition to being faster and usually less expensive, SAS drives, although smaller, allow for

more spindles in the same rack space.

Regardless of the technology used, a consideration when designing RAIDs is that it is generally best to

utilize multiple channels or ports for high-throughput logical drives. For example, an 8-drive RAID 1+0 on

a single U320 channel provides 320MB/s of sustained throughput, while the same drives in a RAID with

four on each channel of a two-channel U320 controller (striped within the channels and mirrored between

channels) provides 640MB/s sustained throughput and offers additional fault tolerance to controller

channel or cable problems. If each quad of drives is in a different cabinet, this setup also provides fault

tolerance for cabinet failures.

An alternative to local storage for VCM is to use SAN storage. A common problem with SANs and VCM

previously was that many SANs are designed for file server or mailbox use and are not well-suited to

high-throughput OLTP-type activities. For a SAN to provide good performance with VCM, it must be

properly configured internally, and all devices between the SAN and the Collector must be adequate for

the task. A 4Gb HBA is capable of slightly higher throughput than the single Ultra 320 SCSI channel. For

write activities, since mirroring and striping is handled internally at the SAN, the throughput of the 4Gb

HBA is more comparable to two and a half U320 channels. Getting that throughput depends also on the

switches and links between the Collector and the SAN and between the drives and the controllers in the

SAN.

When considering SAN storage for VCM, throughput (read and write speed) and access latency should be

considered. Both are important because VCM performs many relatively small reads and writes. If the

latency is too high, performance will be impacted as SQL Server waits for responses to these small queries

before it can perform the next task.

After you have installed a Collector, analyze performance of the disk subsystem using Performance

Monitor. The main counters of interest are the Physical Disk object’s Disk Bytes/sec and Average Disk

Queue Length counters. You can monitor both of these counters on a per-instance basis to determine the

throughput and the number of threads that are queued for each logical drive that is associated with VCM

activity.

The Disk Queue Length value is the best initial indicator about whether a logical drive has sufficient

throughput and access speed for the tasks being required. Typically, the Disk Queue Length should not be

more than twice the number of processors in the system for more than very short periods of time. When

viewing this counter, a logical drive that is also used by the page file may show high queuing due to

insufficient RAM in the system, but this counter can be very useful in determining whether disk subsystem

resources are appropriate and whether the resources are optimally arranged, such as disks per channel,

RAID type, etc.

Configure SQL Server for VCM

VMware, Inc. 51