How Do I Boost Response Time?

Simply adding storage capacity is rarely a successful solution for disk response time issues. Most often, response time problems are caused by volume configuration issues and lack of bandwidth. However, in some cases, adding additional disks can improve response time due to the additional spindle count.

To fully address response time issues, the storage administrator must work with server and application administrators to fully assess the situation. The first goal should be to quantify the issue, and the most logical first step is to leverage server-based disk performance monitoring tools to determine the current state baseline for response time metrics.

Next, the storage administrator should test the capabilities of the current infrastructure. By doing this, the administrator can pinpoint the cause of problems -- or opportunities for improvement. Response time issues can originate at any point in the data path, and storage isn't always the bottleneck. However, intelligent storage arrays can provide a significant opportunity for response time improvement.

Once a full situation assessment is complete and the storage infrastructure is determined to be the target of improvement opportunities, there are several areas that should be explored:

Data transfer mechanisms can offer significant opportunities or problems for disk response time. The HBA cards, connective cables, switches, and storage ports are the gateways for data movement and SCSI command completions.The key metrics for measurement in this area are bandwidth utilization and hop count. Both of these metrics should be monitored for possible configuration improvements. Typical effective practices dictate a maximum of 80 percent peak bandwidth utilization and no more than two hops (connection and transfer points) between the server and its storage.

Storage array cache is often the key to dramatic increases in response time. Intelligent cached disk arrays can circumvent the traditional delay caused by the physics associated with writing data to spinning disks. Cached disk arrays can provide the equivalent of solid state disk for disk write operations if sized and configured properly. Likewise, many applications that require continuous sequential reads can achieve near solid state response times. Key metrics in this area include cache utilization and percentage of read cache hits. Cache utilization can be mitigated by adding more memory, but read cache hits can only be improved by changing the application's read sequencing.

RAID levels can also affect response times. Parity calculations required for some RAID configurations can adversely affect response times for highly random read and write environments. RAID types should be carefully researched and compared to the type of application environment to maximize response time performance.

As mentioned earlier, in a SAN environment, most application storage can be optimized for response time by leveraging cache. However, some cache unfriendly applications (Exchange and some SQL instances are good examples), need maximum spindle count to optimize response time. Many application vendors publish best practices or calculators to determine optimum spindle counts for specific application configurations.

Disk performance tuning is usually tricky and usually referred to as an "art form." The response time performance vertical is no different; there are an abundance of possible tweaks and changes that should be considered when delving into the world of response time tuning. Storage administrators should take a systematic approach and utilize all of the benefits of enterprise storage to optimize overall application performance and availability.Tim Arland, Principal Consultant for Storage Solutions, Forsythe Solutions Group Inc.