SAS is constantly evaluating new technologies for internal and customer experiences. Based on requests from many of our customers, SAS R&D engaged with Fusion-io® to do performance testing with Fusion-io® ION™ Accelerator and ioScale™ storage devices. Our testing goals for the testing included:
- Garnishing a relative measure of how Fusion-io® storage performed with SAS workloads
- Determining if flash storage would yield substantial benefits for SAS large-block, sequential IO pattern
- Get a relative comparison of Fusion-io® flash with traditional spinning disk
- Discover any tuning of the devices or host that could improve flash performance, versus “out-of-the-box” experience
Our test bed was a mixed analytic workload composed of heavy IO, computational, and memory intensive tests. The matrix test bed was designed for a heavy concurrent mix of long and short jobs, averaging 30 concurrent jobs, and jobs stacked end-to-end. This test bed is a good representation of the average job mix many SAS shops have employed.
The test environment was a 4-socket, HP Proliant DL980 with a total of 64 cores, 512GB RAM. The Fusion-io® environments we tested were in-box direct acceleration with 4x3.2TB ioScale and 8GB FC attached ION Data Accelerator with 6x3.2TB. Testing involved 120 simultaneous concurrent jobs as this offers a good workload that stresses system resources of CPU and system throughput.
In summary, the Fusion-io® flash storage performed admirably. Both the ION™ Accelerator and the ioScale™ devices (with slightly varying results) reduced the aggregate elapsed time of all 102 jobs by over 60 percent compared to a spinning disk array we have attached to the same computer system. This was done by significantly reducing IO latency to the SAS application.
Initial results “out-of-the-box” didn’t perform nearly as well as our final tests after tuning the Fusion storage. We worked with Fusion-io engineers to tune host kernel, processor, device, file system and other settings to maximize the flash performance, and it made a very significant difference.
In conclusion, our testing experiences show that Fusion-io® flash storage can be very beneficial and cost effective for SAS workloads. Every workload is different, and it is important to understand your existing performance needs in terms of CPU, network, and storage I/O. Fusion-io® can accelerate storage I/O in terms of reduced latency and increased throughput, and it is recommended your work with your Fusion-io® engineer to maximize product, deployment architecture, and system tuning to achieve the best result. You can review detailed reports for each storage device:
- Performance and Tuning Considerations for SAS® on Fusion-io® ION Accelerator™
- Performance and Tuning Considerations for SAS® on Fusion-io® ioScale™ Flash Storage
Does the workload need to be at such a level to require a machine (64/512GB) like the one used in their testing before such an IO sub-system is beneficial?
Not really. The need to have a SAS workload (one job or multiple) complete in a certain timeframe is all this is required to support this type of IO subsystem beneficial.