SPC Engineer Swallows the Automated Performance Tuning Pill

Those of us who cut our teeth on single-disk storage systems and were heavily involved in performance optimization often find it difficult to take the leap of faith required to trust many new technologies. I (James) can remember when RAID came out. At the time, I was working on Oracle databases and wondered why I needed RAID and if it would stand the test of time. After all, many database professionals were already manually striping data across discrete disks. With RAID technology we had to overcome our inherent need to control data placement and embrace the ease of deployment and gains in performance, reliability, and availability.

Trusting technology to perform the tasks administrators and engineers were once asked to do can be a tough pill for these hard-core folks to swallow. Especially so because we tend to know a lot about the systems we manage (workloads, I/O patterns, utilizations, etc.). Yet sometimes we have to let go of our natural human tendency to use manual processes and let automation take over.

One of the more powerful stories Jerome and I have heard recently in this regards involved a performance engineer at 3PAR preparing for the company's first-ever Storage Performance Council (SPC) benchmark back in 2003. This engineer believed he needed to adjust 3PAR's default virtual volume layouts in order to achieve optimal performance results. This engineer literally spent weeks tuning the 3PAR system by trying to manually place volumes on the 3PAR system in order to achieve the best possible SPC performance test results.

Realizing customers would not want to nor would have the time to tune its storage systems, 3PAR's VP of Engineering insisted the SPC test be redone with 3PAR's default configuration. Expecting sub-standard results, the engineer grudgingly performed the tests. But to the engineer's amazement the results from the default configuration were remarkably better than the configuration he had so tediously labored over on for weeks.

What this helped to prove is that even using a knowledgeable expert with years of experience in tuning storage systems can now be outdone by the features now natively found in the 3PAR system. It also goes to prove the point that todays distributed system environments are too dynamic for any one individual to keep up with, even highly trained and experienced engineers.

Yet this is what corporations should expect from todays enterprise storage systems to deliver - automated self-tuning capabilities that stand up to even to the most rigorous tests such as the SPC's performance benchmark tests. 3PAR's approach to dynamic optimization and thin provisioning and its results in the SPC test demonstrate how performance tuning is now becoming a function that is best left to the storage systems themselves.

To do this, 3PAR storage systems provide a 3-level mapping methodology that first puts all physical disks into a large virtualized pool of storage that are divided into uniform-sized chunklets of 256 MB in size. 3PAR then creates virtual volumes from these chunklets with each virtual volume consisting of chunklets that are spread across all the physical drives in the 3PAR storage system. 3PAR then further optimizes this configuration through the use of thin provisioning so space on a disk drive is only consumed when data actually exists. By creating wide-striped volumes across all array resources (disks, controllers, cache, ports) coupled with 3PAR's mapping technology helps to address environments where a high level of random I/O exists across a system and helps to improve the number of IOPS (I/Os per second) delivered by the system.

Hard-core administrators and engineers may find it a tough pill to swallow that computers can now do a better job of performance tuning and data distribution than they can (by default, no less) but as 3PAR's recent SPC-1 performance test that reached over 224,000 IOPS demonstrated, that day is now upon us. The 3PAR storage system provides companies the flexibility to load balance, thinly provision and optimize performance on its system in such a way that it meets the demands of distributed environments with various workloads without requiring them to bring in specialists or dedicate their own people to spending weeks to configuring the system. However as any administrator or engineer who is honest with themselves know, today's workloads are just too difficult to keep up with and optimize and it will only make and your organization look better is you use a storage system that can keep pace with the change within it.