Hi Greg, we met each other recently at the Dutch Storage Expo after one of your sessions. We briefly discussed the current trends in the storage market, and the "risks" or "threats" (read: challenges) it means to "us", the storage guys. Often neglected by the sales guys...
Please allow me a few lines to elaborate a bit more and share some thoughts from the field. :-)
1. Bigger is not better?
Each iteration in the new disk technologies (SATA or SAS) means we get less IOPS for the bucks. Pound for pound that is. Of course the absolute amount of IOPS we can get from a HDD increases all the time. where 175 IOPS was top speed a few years ago, we sometimes see figures close to 220 IOPS per physical drive now. This looks good in the brochure, just as the increased capacity does. However, what the brochure doesn't tell us that if we look at the IOPS/capacity ratio, we're walking backwards. a few years ago we could easily sell over 1000 IOPS/TB. Currently we can't anymore. We're happy to reach 500 IOPS/TB. I know this has always been like that. However with the introduction of SATA in the enterprise storage world, I feel things have gotten even worse.
2. But how about SSD's then?
True and agree. In the world of HDD's growing bigger and bigger, we actually need SSD's, and this technology is the way forward in an IOPS perspective. SSD's have a great future ahead of them (despite being with us already for some time). I do doubt that at the moment SSD's already have the economical ability to fill the gap though. They offer many of thousands of IOPS, and for dedicated high-end solutions they offer what we weren't able to deliver for decades. More IOPS than you need! But what about the "1000 IOPS/TB" market? Let's call it the middle market.
3. SSD's as a lubricant?
You must have heard every vendor about Adaptive Storage Tiering, Auto Tiering etc. All based on the theorem that most of our IO's come from a relative small disk section. Thus we can improve the total performance of our array by only adding a few percent of SSD. Smart technology identifies the hot tracks on our disks, and promotes these to SSD's. We can even demote cold tracks to big SATA drives. Think green, think ecological footprint, etc. For many applications this works well. Regular Windows server, file servers, VMWare ESX server actually seems to like adaptive storage tiering ,and I think I know why, a positive tradeoff of using VMDK's. (I might share a few lines about FAST VP do's and dont's next time if you don't mind)
4. How about the middle market them you might ask? or, SSD's as a band-aid?
For the middle market, the above developments is sort of disaster. Think SAP running on Sun Solaris, think the average Microsoft SQL Server, think Oracle databases. These are the typical applications that need "middle market" IOPS. Many of these applications have a freakish IO pattern. OLTP during daytime, backup in the evening and batch jobs at night. Not to mention end of month runs, DTA (Dev-Test-Acceptance) streets that sleep for two weeks or are constantly upgraded or restored. These applications hardly benefit from "smart technologies". The IO behavior is too random, too unpredictable leading to saturated SATA pools, and EFD's that are hardly doing more IO's than the FC drives they're supposed to relief. Add more SSD's we're told. Use less SATA we're told. but it hardly works. Recently we acquired a few new Vmax arrays without EFD or FASTVP, for the sole purpose of hosting these typical middle market applications. Affordable, predictable performance. But then again, our existing Vmax 20k had full size 600GB 15rpm drives, with the Vmax 40k we're "encouraged" to use small form factor 600GB 10krpm drives. Again a small step backwards?
5. The storage tiering debacle.
Last but not least, some words I'd like to share with you about storage tiering. We're encouraged (again) to sell storage in different tiers. Makes sense. To some extent it does yes. Host you most IO eager application on expensive, SSD based storage. And host your DTA or other less business critical application on FC or SATA quality HDD's. But what if the less business critical application needs to be backed up in the evening, and while doing so completely saturates your SATA pool? Or what if the Dev server creates just as many IO's as the Prod environment does? People don't seem to care it seems. To have people realize how much IO's they actually need and use, we are reporting IO graphs for all servers in our environment. Our tiering model is based on IOPS/TB and IO response time.
Tier X would be expensive, offering 800 IOPS/TB @ avg 10ms
Tier Y would be the cheaper option offering 400 IOPS/TB @ avg 15 ms
The next step will be to implement front end controls an actually limit a host to some ceiling. for instance, 2 times the limit described in the tier description. thus allowing for peak loads and backups.
Do we need to? I think so...
Greg, this small message is slowly turning into a plea. And that is actually what it is, a plea to our storage vendors, and to our evangelists. If they want us to deliver, I feel they should talk to us, and listen to us (and you!).
ps, I love my job, this world and my role to translate promises and demands into solutions that work for my customers. I do take care though not to create solution that will not work, despite what the brochure said.
pps, please feel free to share the above if needed.