On friday @plankers sent this Tweet:” 2.5″ 10k disk is as fast as a 3.5″ 15K disk. Did not know that, thanks @Compellent #TechFieldDay“! yes, it sound a little bit strange and it is inaccurate but twitter’s 140 chars limit doesn’t help to fully explain the whole story.
Simply put, this result is achieved positioning most accessed data on the outer tracks of the disks but, after this Tweet, a furious quarrel started about doubts some competing vendors raised. The discussion was closed by the storage anarchist and alextangent with twitts like: “the math (& observed reality) doesn’t support the mktg BS. BTW the BEST data placement is MIDDLE of platter” and “mid-platter data placement well understood since winchester tech 1st appeared; 1970-80s!” …. But it’s not finished yet!
Well, they tried to do some unrealistic FUD stating that Compellent’s Fast track is based on wrong concepts and it doesn’t work!
A short prologue
Compellent has a software feature called Fast Track. This features is used, with success, by many Compellent customers to get more IOPS per disk! It works and satisfied customers can confirm this.
I wrote a lot on FT in the past and you can find articles in my blog by clicking here, here, here or searching “fast track” for a full list.
You can also find a nice and illuminating video about Fast Track on the Compellent’s website.
How is it possible?
It’ simple Compellent Storage Center is a next generation storage system and was architected less then 10 years ago (not in 80s o even in 70s!).
The Compellent architecture is not comparable to old generation (i.e.: clariion or V-Max) arrays because it redefines the basic concepts about how data blocks are stored and managed! Comparing Compellent’s storage architecture to the EMC’s arrays architecture equals to compare Technology and Archeology!
When Barry and Alex say that they get the best performance from the array positioning most accessed data blocks on the middle of the disk they speak about technologies of the last century: (winchester disks: 40 years ago!), arrays (designed on archaic raid techniques) and very old algorithms (those you’ll likely find in the base foundation of old generation arrays)… so we must admit we are speaking about archeology surely not about nowdays technology!
The options are two: These guys are biased or they ignore how Compellent works (probably both).
Fast track works because it is integrated with data progression: a true Fully Automated Storage Tiering available since years, not a feature like EMC’s FASTv2 promised more than 1.5 years ago! You can’t expect automatic data placement optimization (like fast track) from companies that can’t offer a true Automtated Tiered Storage.
In vitro Vs. real life
We have studied and tested all the Compellent software features intensively before start to sell these systems in Italy two years ago and I can say that fast track works: we have a lot of documentation on this and we achieved more than 30% improvements after having enabled it on our system.
The world is not perfect and 30%, sometimes, drops down to 20 or even 15: it depends on the number of the hosts and the applications involved but this anyway means that you can buy up to 20% less disks!
Now, if I add my 20% to a 10/12% of 2,5″ Vs. 3,5″ (improvement in IOPS due to the less latency), I get more or less 30/32% improvement. That’s why a 10K/2,5″ disks “equals” a 15K/3,5″ IOPS!
Obviously I’m not telling you a 2.5″ 10k disk is as fast as a 3.5″ 15K disk. But we can say that the fastest tracks of a 10K RPM 2,5″ disk can give you the same average IOPS a 15K RPM 3,5″ disk can offer… If you can use them the right way.
Last, but not least, Barry where do EMC arrays write the most accessed data blocks? I hope automagically in the middle of the disk! Because a Compellent’s array can decide where to put most accessed data but I’m not sure Clariions and VMaxes can!