IOPS and bandwidth

There are so many factors that go into the throughput and bandwidth numbers that I generally ignore the marketing claims of the "theoretical capabilities" for entire arrays. What it really comes down to is determining what you need and figuring out if you can meet those needs with a specific configuration. Generally to actually achieve the "theoretical max" available from an array you have to have an absolutely ideal lab environment.

Instead you need to figure out what you need per host/application, determine how these hosts are going to attach and then work your way up to figuring out if you can achieve your requirements with a specific model or if you need to look at something bigger. Don't forget that other factors enter in to the equation as well... like your HBAs (how many and what speed, and does the server bus support their full speed), SAN switches (port speed, overall load, blocking or non-blocking technology depending on overall throughput requirements), and even the quality of the cables you use (Don't try to run over 50 meters with an OM2 cable at 8G FC).  Yes, this all sounds really picky, but but it is just supporting the statement that the overall limits published for arrays are really only good for general comparison under ideal conditions. They give you a feel for how the relative performance will improve (or be hampered) but they don't really tell you a whole lot else.

Mind you, this is just my opinion. I'm sure there are other opinions out there :-)

Hi Allen,

Agree with you regarding marketing info.

I can easely see how much my current apps are consuming in terms of IOPS and bandwidth with Navisphere Analyzer. I have apps asking to migrate from other vendor arrays to this one since it's newer technology. How can I really know there is still room and how much?

I just don't want to beleive it will be ok. I want to measure it and be sure.. I know how many IOPS and bandwidth the new app will consume

Thanks,

Pablo

Have you looked at the two Best Practice guides in the Documents section? One is overall general and one is specific to flare release number. If you use Analyzer, you can see what the overall Utilization for the SP's and there are best practice recommendations based on Utilization and queue lengths. If the SP's have low utilization, say under 30%, that would mean that you could add more load.

It's also based on the number of disks you currently already have installed - how many more can you add - this will determine what you can normally add. Use the methods in the Sizing section in the Flare Best Practice guide to get an approximate size that you could add.

This is a really hard question to answer as there are some many variables - you might want to contact the local EMC Professional Services group for help.

glen

The IOps and bandwidth for an array is mainly dependent on the drives attached to the array.

a 15k rpm FC drive can do about 180 IOPS.

a sata drive can handle about 80 IOPS.

You can use navi analyzer to find out how much IO each drive is doing.

so if you have a 4+1 raid 5 raid group of 15k rpm FC drives currently doing 500 IOs in total.

you can workout the maximum 5*180 = 900

So you know that if you add an application the is doing 400 IOs or less you should be ok.

Thanks Brian for your answer. That explain my question partially. I understand now that we have to think about IOPS on a RAID group.

Now:

1) If I have a RAID 5 Thin Pool with 50 15K rpm drives, how many IOPS it can handle?

2) If I have a Thin Pool of 25 SSD drive, how many IOPS it can handle?

3) A single application can easly have 10 LUNS, how can I know how many IOPS the application is doing? Not LUN by LUN.

Thanks in advance,

Pablo

The answers to 1-2 will depend on how you provision the Pool.  The virtual provisioning wizard will take the number of drives and create RAID 5 or RAID 6 groups based on specified rules.  For example, 20 drives provisioning a RAID 5 pool will result in 4x (4+1) RAID groups created 'under the hood' in the Pool.  You can calculate the Pool IOPS in the same way you calculate FLARE IOPS based on the RAID group size.

This is described in greater detail (including the rules) in EMC CLARiiON Best Practices for Performance and Availability FLARE Revision 29.0in the Virtual Provisioning section.  This document is available on Powerlink.

For question 3 you'll need to ask your DBA or System Architect before hand to make an estimate, or you can measure the IOPS at the CLARiiON's ports using Analyzer, if the application is implimented.