Well, That Was Amusing
Just got off the webcast that Intel, Adaptec and the Burton Group did on FlashSSD (Intel’s X25-E Extreme) and Adaptec’s shiny new MaxIQ controller.
They presented a quadrant chart bounded by risk and CAPEX/OPEX cost. In the low risk/high cost quadrant, they positioned deployment option 1: just replace all disk with Flash SSD. Higher risk with high cost was option 2: doing a mix of FlashSSD and disk customized to support specific apps. They said this approach is “tricky” since you may have to rewrite apps to take advantage of the SSD (plus it may screw with your virtual servers). Option 3 (lower risk/lower cost): Use the Intel and Adaptec products together and let the Adaptec card figure out what data should be put into the Flash SSD and what data should be written to your disk using Adaptec’s review of I/O en route to the back-end storage. They didn’t populate the quadrant for high risk/low cost — perhaps this is the HDD-only category.
They spent most of the Q&A session discussing (writing off, really) the notion that memory wear was a gating factor in cost-efficiency and performance. Their statements were general and unconvincing. Flash SSD memory wear algorithms on the Intel stuff make the technology good enough. So, just faggetaboutit!
There was no discussion of any monitoring that would let you track memory wear or predict the frequency with which you might need to swap out the Intel Flash SSDs. Do you simply wait until performance degradation increases?
They also said that time would tell you whether you might need a more expensive Multi Layer Cell Flash SSD verses a Single Layer Cell unit (which costs less).
There was no discussion of what the hold back is on total Flash SSD capacity to support wear leveling and cell substitution. Are you using what you buy, or only a fraction of it?
A comparison was made between a 2u 7 drive SAS array with an Adaptec controller and a 1u 3 drive SATA array plus 1 Intel Flash SSD and a MaxIQ controller: the latter improved transaction rates by 5-10x, provided 50% greater storage capacity (900 GB at 2000 IOPS versus 1.5 TB at 20,000 IOPS), used half the power and half the rack space. Interesting comparison, but it begs the question of whether I want to use high capacity, lower cost SATA front-ended by prone-to-wear Flash SSD versus enterprise class SAS…
Lots of references to current woo about server virtualization, clouds and de-duplication, but I missed any real connection between these technologies and the advantages of the Intel/Adaptec solution. Perhaps they were arguing that we can brute force a solution to VMware’s “brain-deadness” when it comes to I/O handling.
Of course, Intel argued that CPU speeds have increased 175x while storage speeds have increased by only 1.3x in the same timeframe. This is an “I/O gap” that creates storage bottlenecks. I realize the gap exists and that it always has. Someone smart once said that processor, transport and storage were always playing hop scotch.
If you have about 45 minutes to kill, you can watch the replay at ITOAmerica.com.