FalconStor And Violin Add SSD To NSS
While I had been waiting for FalconStor to add flash support to their Network Storage Server (NSS) storage virtualization software, I was expecting flash volumes off a Fusion-IO or TMS PCIe flash card with promises of automated tiering to arrive sometime before Snow White's prince. I was pleasantly surprised when the folks at FalconStor called to tell me they were aiming a little higher than that and using Violin's solid state memory array as a cache.
March 10, 2010
While I had been waiting for FalconStor to add flash support to their Network Storage Server (NSS) storage virtualization software, I was expecting flash volumes off a Fusion-IO or TMS PCIe flash card with promises of automated tiering to arrive sometime before Snow White's prince. I was pleasantly surprised when the folks at FalconStor called to tell me they were aiming a little higher than that and using Violin's solid state memory array as a cache.
Frequently used to turn an industry-standard x64 server and attached JBODs into a SAN array, FalconStor's NSS, formerly known as IPStor, can do much more than that. Not just a software target for any mix of iSCSI, Fibre Channel and Infiniband. NSS turns a server into a full-featured block storage controller providing value added services like snapshots, data migration, multipath support and heterogeneous replication using any storage from a SATA Jbod to an enterprise array.
The Violin Memory 1010 is a rack mount appliance that can hold up four usable terabytes of NAND in 84 flash modules. Violin claims this architecture allows them to both get greater utilization of the flash than the usual mirrored pair of Zeus SSDs with RAID parity and process over 220K write IOPS with an admirable 20-70 microseconds of latency. Test data they showed me indicates that it can sustain the 220K IOP rate even when being stressed by a benchmark application for hours.
Connect a Violin 1010 to an NSS server using a PCIe extension cable to provide maximum bandwidth and the NSS server will accelerate access to volumes you select by caching both reads and writes This isn't plain-old, least-recently-used block-caching, oh no. The NSS keeps block access frequency metadata. It uses this access temperature data and the user assigned LUN priority to calculate what data to cache in real time.
FalconStor's benchmarks indicate that an SSD cache of one percent the size the data behind it can improve performance of applications like ERP. Starting at $32,000 with 500GB of flash it should be a better solution than slapping a few STEC ZUES SSDs in the old disk array.
About the Author
You May Also Like