Dataram today took the wraps off its XcelaSAN solid-state disk (SSD) appliance. The company made its name selling memory-optimization products for high-performance computing workloads (and claims 70 percent of the Fortune 100 as customers), but the revolutionary promise of SSD technology lured Dataram into the storage arena. Even as a storage newcomer, the company is banking on its chances for success, citing its intelligent data-analyzing software as a major differentiator between its appliance and other SSD products.
The XcelaSAN appliance sits between the Fibre Channel switches and the disk arrays, providing a 128GB cache to serve a company’s most-active data. However, unlike most SSD products, which require users to decide which data can benefit most from the fast performance, Dataram’s proprietary algorithm handles that task automatically (and continuously). Chief Technologist Jason Caulkins said that XcelaSAN starts determining which data to serve from the cache as soon as it is plugged in, and can complete the process in about an hour. Once it is up and running, an XcelaSAN appliance can handle 450,000 IOPS at 3GB per second. The appliances cost $65,000 apiece.
That price tag is a bit high compared with hard disk drives, but it is important to remember that in addition to actually boosting performance, SSDs eliminate the hardware costs associated with increasing storage-area network (SAN) performance. Instead of keeping data across the outermost areas of many spinning disks to improve performance (a process known as “short stroking”), SSD users can put high-performance data in the cache and more fully utilize a far lesser number of storage arrays for the rest. According to Caulkins, XcelaSAN would allow a company managing a 1TB dataset to reduce its number of SAS disks down to 25 from 200.
He also pointed me to a comparison with the Transaction Processing Performance Council’s top system for price performance. Dataram contends that, all else being equal (server, database, etc.), a business using XcelaSAN to handle the same workload could reduce operational expenditures by two-thirds; reduce capital expenditures to $104,000 from $124,000, for example; and increase transactions per minute (TPM) to 500,000 from 232,000, while reducing the cost per TPM by 64 percent. “All of a sudden, that dollars-per-gigabyte discussion is not the important discussion,” Caulkins explained. “It’s the dollars-per-performance discussion [that is important].”