Wide striping is a two edged sword

wyhlg

当XIV的sales 对你的忽悠说，XIV可以提供和FC阵列提供一样的IOPS时，千万要记住，这是在牺牲存储的利用率的基础上得到的。
I have spent a lot of time lately talking with some
of my coworkers, friends, etc. on the topic of wide striping. This
topic keeps coming up since there are now a number of vendors selling
storage arrays with SATA drives that claim to have "the same
performance as fiber channel". Some of the Sales folks I work with keep
asking how we are supposed to dissuade people from that idea, or if
it's true. One of the prime offenders in this regard is IBM with their
new XIV array. The XIV uses wide striping and SATA drives and they
claim to have "enterprise performance" at a very low price point. But
they aren't the only ones; you have Dell telling people the same thing
about their EqualLogic line of storage as well, and there are other
too. For an excellent article about the XIV and its performance claims,
take a look at
http://thestorageanarchist.typepad.com/weblog/2009/01/1037-xiv-does-hitachi-math-with-roman-numbers.html
.
What
I usually tell them is that the statement is true; you can get fiber
channel performance by striping across a large number of SATA drives.
The only problem is that you have to give up a lot of usable disk space
in order to keep it that way. A quick example usually illustrates the
point quite well. Let's say that for the sake of easy math the average
application in your environment uses about 5TB of space (I'm sure some
are a lot more, and some a lot less, but we are talking average here).
Let's also say that you need about 2,000 IOPS per application in order
to maintain the 20ms max response time you need in order to meet the
SLAs you have with your customers. Finally, let's also assume that your
SATA array has about 90TB of useable space using 180 750GB SATA drives
and you can get about 20,000 IOPS in total from the array. So, let's do
some basic math here. That means that you can run about 10 applications
at 5 TB apiece which will take up about 50TB. So, your array will
perform well, right up until you cross the ½ full barrier. After that,
performance will slowly decline as you add more application/data to the
array.
So, what does this mean? It means that the cost per GB
of these arrays is really about twice what the vendors would have you
believe. OK, but considering how much cheaper SATA drives are than 15K
fiber channel drives, that's still OK, right? Sure, as long as you are
willing to run your XIV at ½ capacity. In today's' economic climate,
that's going to be tough to do. I can just imagine the conversation
between your typical CIO and his Storage Manager.
Storage Manager – "I need to buy some more disk space."
CIO
– "What are you talking about, you're only at 50% used in theses
capacity reports you send me and we didn't budget for a storage
expansion in the first year after purchase!"
Storage Manager
– "Well, you know all that money we are saving by using SATA drives?
Well, it means I can't fill up the array; I have to add space once I
reach 50% or performance will suffer."
CIO – "So let
performance suffer! We don't have budget for more disk this year. Why
didn't you tell me this when you came to me with that 'great idea' of
replacing our 'enterprise' arrays with a XIV?!?!"
Storage
Manager – "Ahhh … ummmmm … gee, I didn't know, IBM didn't tell me! But
we had some performance issues early on, and figured this out. Do you
really want to tell the SAP folks that their response time is going to
double over the next year?"
CIO – "WHAT! We can't let that
happen, we have an SLA with the SAP folks and my bonus is tied to
keeping our SLAs! How could you let something like this happen! Maybe I
should use the money for your raise to pay for the disks!"
Storage Manager – "Um, well, actually, we need to buy an entire new XIV, the one we have is already full."
OK,
enough fun, you get the idea … make sure you understand what wide
striping really buys you and if you decide that the TCO and ROI make
sense, make sure you communicate that up the management tree in the
clearest possible terms. Look at the applications that you currently
run, see how much space they require, but don't base the sizing of your
EqualLogic (see, I'm not just bashing the XIV) just on your space
requirements. Base them more on your IOPS requirements. With SATA
drives chances are pretty good that if you size for IOPS, you'll have
more than enough space.
               
               
               

本文来自ChinaUnix博客，如果查看原文请点：http://blog.chinaunix.net/u2/75575/showart_1891160.html