Last week I was listening to a HP video where the HP CTO said that Fibre Channel drives will cease to exist in the coming years. Instead, enterprise storage disk arrays will be configured with two tiers of storage: SSD and SATA. I thought for a minute, and this made sense. EMC is also spelling doom for FC drives.
SSDs are super fast, lower power, high IOPS, but very pricey today. SATA is “slow”, cheap but fairly reliable. Fibre Channel drives are certainly faster than SATA, but not orders of magnitude faster. The trick is putting applications that require high IOPS on the SSD, but less performance oriented apps on your SATA storage.
The key to the whole concept is minimizing use of SSDs, maximizing SATA use, yet still achieve application SLAs. Last year when I was shopping for an enterprise array Compellent touted their sub-LUN level optimization. Blocks of data that are used frequently are left on more expensive tier-1 disks, while less used data is migrated to slower, cheaper disks. This is sub-LUN level, meaning different pieces of your Exchange database could be on two or more tiers of disks at the same time with automatic data migration. But I wouldn’t put Compellent in the catagory of true enterprise class storage arrays.
Last year EMC announced that a future version of FAST (Fully Automated Storage Tiering) would also work at the sub-LUN level. Their currently shipping version of FAST only operates on entire LUNs. I’ve heard rumors that FAST v2 may ship sometime later in 2010.
As I was reading my dozens of RSS feeds today, I noticed that 3PAR announced they will be supporting STEC enterprise SSDs and sub-LUN level automated storage tiering. Their adapative optimization (AO) feature is orderable immediately, but is unclear when their arrays will actually support it. A feature I found really interesting is their QoS gradient concept. You can bias the automated data migration based on performance or cost objectives. So you could have one LUN that needs to be performance oriented, but another that was more cost sensitive.
3PAR claims to be the first enterprise storage company with sub-LUN level automated optimization. If 3PAR can ship AO before IBM, HDS, HP, NetApp, and EMC then, yes, they will have delivered another industry leading feature to enterprise customers before the ‘big boys.’ I don’t know how much AO would help in a Fibre Channel and SATA only environment, since there’s not a huge performance gap between the tiers. Inquiring minds want to know!
I hope 3PAR can develop some type of ‘sizing’ tool for SSDs and AO that you can run against an existing array to baseline the performance characteristics over a period of time (days or weeks), then provide recommendations on how much SSD storage is recommended to meet certain performance SLAs.