Data Doesn't Move at a Constant Speed ...

Have you sat and people-watched for a period of time? I travel quite a bit and sometimes I like to sit in an airport and watch people as they move back and forth in the airport. The one thing I've noticed is that people move at different speeds. Some move really fast (like those late for a flight), and some move really slow (like they have lots of time on their hands -- like me sometimes). People-watching is also fun to do in large tourist areas such as Las Vegas or London. If you happen to be in a tourist area, watch the behavior of large groups -- lots of fun!

I was talking to an HPC user recently, and we were discussing the need for having storage with different performance characteristics. I was reminded of the similarity of the discussion to my people watching. Sometimes data needs to have a great of performance in terms of throughput and IOPS (i.e., it needs to be fast like it's late for a flight) when applications need the data to keep processing (this is true for either reading or writing). But then, after a period of time, the data doesn't have to move quickly (i.e., low throughput and low IOPS, like travelers who have a great deal of time).

The fun question is how to combine really fast storage and slow storage into a single solution. You'll never get it into a single box despite what vendors tell you. Ideally, you want the data to be where you need it, when you need it, and with enough performance so that your code is not bottlenecked by throughput or IOPS. I love this idea, but I haven't seen anyone successfully do even a small bit of it in an automatic manner. You can do some of it in a manual manner, moving the data from one storage tier to another by hand such as with a script. But I digress -- this is a discussion for another day.

Truly understanding that your data doesn't need to be accessed with the same performance is one of the keys to advancing in your understanding of storage. If you understand this fundamental idea, then you can begin to better architect your storage solution and save lots of money in the process. (As a side note, I use the word storage solution to include hardware and software as well as tools. It does not mean hardware by itself.) I am continually shocked to listen to people discuss storage, and it becomes painfully obvious they don't understand this idea including storage pundits and storage experts. Do yourself and your customers a HUGE favor -- take this message to heart. A famous person once said, "Learn it. Know it. Live it."

Labels: data management,speed,Storage Performance,Storage

posted by: Jeff Layton