Opinion

posted on 20 June 2008 17:16

Here's something for the weekend: it didn't take long. Just one day after HP announced its server flash SSD caching with Fusion-io's ioDrive and a thought that soon we would be having discussions about where best to put flash SSDs; in the server as cache or the storage array as tier zero, up springs Chuck Hollis, EMC's terrific blogging VP for Technology Alliances and jumps straight in.

Chuck (pictured left) quotes TechTarget's Beth Pariseau quoting HP's director of marketing at Storage Works, Patrick Eitenbichler, who says: "I've been very entertained by what other vendors have put out there about solid-state drives. We test devices from all the solid-state vendors in our labs, and we break every one of them. As Tier 1 storage, it wears out very quickly. It's not durable, and the cost benefit is not there." Naturally Chuck is highly entertained by HP then using this unreliable technology in its servers - and so are we all.

My information from HP is that it is "actively investigating how to use SSD's and how to implement them (in EVA arrays) in the future. The falling cost of SSD technology is making it very attractive."

He is keen to point out the disadvantages of putting flash SSD caches in servers. To paraphrase him: "We tell customers that (flash SSD) storage is an expensive, shared resource, and should be pooled -- and used where needed. Don't put it in your servers -- that's a waste. ... The only people who would think this is a really great idea would be server vendors, I guess."

Sun and HP, alsio Dell, are flash caching their servers with Sun going further and changing system SW code, such as ZFS, so that working data is stored in the SSD cache instead of on hard drives in an attached storage resource. HP hasn't said it's going that far but it is early days.

Assuming Sun gets its ZFS software working much faster with a flash cache used in this way then it's surely a good idea. The beauty of it is that it accelerates ZFS operations whether the storage array has a flash SSD tier of storage or not. That's the same type of argument as one might use for InVista, sitting in the SAN fabric and virtualizing storage in SAN-attached arrays, whether or not the storage arrays have their own equivalent storage virtualization or not.

I'd like to suggest an irony here as, arguably, Chuck is taking an HDS USP-like position with regards to flash SSDs being better in networked storage arrays rather than being sited between them and the accessing server's CPU and memory.

One aspect of the server flash SSD caching approach is that physical space constraints may limit the amount of flash cache you can use whereas a separate storage enclosure has lots of hard disk drive slots which can take multiple SSDs. This contrast is probably even more acute with blade servers.

Another aspect of this is that a flash SSD cache for a servers needs to plug in to the server's bus and the supplier doesn't have to worry about getting a Fibre Channel interface onto flash SSDs which is needed to plug them into existing Fibre Channel slots in a storage array. STEC has an effective monopoly on this (with EMC having its own mini-monopoly because of its exclusivity deal with STEC which ends in a few months) until Emulex' SSD-tweaked SATA-to-FC bridge chip becomes available at the end of the year.

If the amount of data you need to give accelerated IOPS is greater than, say, Fusion-io's limit of 320GB - a single ioDrive - and heading past a terabyte then a storage array is probably the better place to locate SSDs from the space point of view. But, given the present cost of enterprise-quality SSDs it looks to me as if it's pretty sensible for server manufacturers to sidestep the STEC/EMC FC SSD interface monopoly trap and accelerate all their attached (non-SSD-using) storage arrays by embedding a flash SSD cache in their servers. This looks to be an excellent starter tactic while they tussle with sticking flash SSDs in their storage arrays, often brought in from Dot Hill, LSI and Xyratex which adds another delay factor to the equation.

Lastly, this tactic also gives the server system vendors' sales reps a response when selling against an EMC Symmetrix bid: "No, our storage arrays are not flash-accelerated - but our servers are and that will accelerate all the storage IOPS too, from the disk tiers."

This is a short-term issue as it's likely that in 12 months or so all high-end, and the high-end of the mid-range, storage arrays will support SSD tiers of storage. STEC has revealed that IBM and HDS will be using its SSDs. 

Perhaps this means that IBM will flash SSD-enable its oft-criticized-by-EMC DS8000 array. That, for lovers of irony, would be terrific. You could just imagine Tony Pearson saying: "Eat your words EMC bloggers one and all."

Have a great weekend.

^{[Chris Mellor.]}