This storage capacity mosh pit Chuck Started

EMC's Chuck Hollis got everybody going last week with his blog about usable capacity for various storage products.   There were lots of interesting things said about it here,      here,     here    and    here    as well as places too numerous to list. 

Pretty noisy, but I think a lot of it missed the point. The focus on usable capacity doesn't get at the heart of the matter, which is how much application data can be safely stored on a given amount of storage capacity without running into performance problems.  Calculating usable capacity is simply configuration algebra.  Let's see it put to work on application data before determining how useful the capacity is!

Vendors educating each other on how to RTFM might be entertaining to some customers, but it's not very helpful to people looking for a solution they can understand.  Why did this convoluted discussion gravitate to who knows how to read which documents the best?  Probably because the operation and configuration of these storage products are far too complicated. 

Also, green field installation scenarios aren't very realistic.  Instead, the discussion should center around how to expand capacity to storage that is already installed and running a mix of applications, such as email, database and file serving.  Assuming a new application comes into this mix, what would the results be for utilization and performance after the new capacity is added and provisioned?

Finally, what advanced storage technologies, such as thin provisioning can be applied to generate realistic, optimized utilization levels?  Of course, this type of discussion requires detailed analysis because there are significant differences between thin provisioning implementations.

Comments

[Stephen Foskett]

I'd still like to hear the 3PAR utilization use case... Come on, guys, you oughta be able to shine here!

[marc farley]

Stephen,

It's an interesting thing, Chuck's blog was actually not about utilization, but usable capacity - and of course there are big differences between the two. (Search Chuck's posts for "utilization" and you'll see what I mean.) Usable capacity that runs out of performance gas at low utilization rates or mixed workloads is not necessarily all that "usable."

So thanks for the prodding, we DO need to provide more information about the superior utilization of our thin provisioning. When you say "use case" what do you mean?

BTW, I'll be out for about a week and I'll pick up on this when I return.

[Stephen Foskett]

Marc,

Rather than post performance numbers, Chuck/EMC proposed a special use case for the storage. They want us to presume that, whatever the SPEC or IOPS, the system they presented could run Exchange and SQL and other enterprise apps at the level of usable storage they suggested. So their use case is a substitute for performance numbers.

By suggesting that x machine can handle x workload at x usable space, they are also suggesting utilization metrics - why have 70% usable if you couldn't fill it up?

What I was getting at is pretty much what your next post shows - that at very high utilization, you get very high spec numbers. Now what was the ratio of usable to raw there?

Later,
Stephen

[marc farley]

The redundancy used was disk mirroring and the ratio of usable capacity to raw capacity was 44%.

Addressable storage was 82,463 GB and total physical capacity was 187,924 GB

As to the question, why have 70% usable if you couldn't fill it up? The answer is performance degradation at higher utilization rates.

Stephen, now I really am going to be gone until mid next week. During that time comments won't get posted and answered. Later.
-marc