Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

1364

November 7th, 2016 08:00

To stripe or not to stripe?

Back in the old days (DMX2/3/4) striping was the way to go if you wanted to use metas. After not having much to do with Symmetrix, I'm now beginning to to work with two VMAX 40Ks again. We have FAST VP and thin devices. I've heard that the new performance rule is to NOT use striped metas anymore and use concatenated metas when creating tdev-metas. I just can't get my head around it why this would be the case.

What do you think is the best way to go for metas? And why?

November 8th, 2016 01:00

Hi dear RRR,

your colleagues may think that backend is already striped in thin provision environment, so no stripe on Meta is needed anymore. It's yes that backend is striped well. And concatenated Meta is easier to be expended than striped one.

But we still can get benefit from striped Meta on performance. There are still some resources such like device queue depth on each Meta member are limited, and striped meta structure may help on burst IO or heavy workload.

You may refer FAST VP Cheatsheet. Refer KB 451282. here there are some topics on meta:

Always use Striped Meta Volumes to create TDEVS if the TDEV capacity is greater than 100GB.

a. Try to limit meta members counts to 16 or less (there are no performance benefits to going wider).

b. However there may be a need to use more members to build larger metas to reach the required lun size (ie. > 4TB), so just use a reasonably large split size).

And also please refer KB 306293, where there are recommendations for using host side striping as well.

Jianyun Yu

November 7th, 2016 17:00

On VMAX3 we don't use meta anymore. but on VMAX40K, striped meta is still preferred for large volumes, no matter it's thin provisioned or not.

2 Intern

 • 

5.7K Posts

November 8th, 2016 00:00

that's exactly what I thought. I just don't get it that 2 of my colleagues were taught to use concatenated instead of striped.

2 Intern

 • 

5.7K Posts

November 9th, 2016 00:00

I'm reading this article: On a VMAX, where is this information : Disk array RAID stripe size

  • The RAID stripe size on DMX and VMAX 20k are all 256K.
  • For the 40K and the new 10K it is 128K for 7+1 and RAID6.  3+1 is still 256k.  It is not adjustable.

So what happens is data is being written in the pool? FAST VP uses 768kB, but striping uses 256kB.

  • Will 3 x 256kB be written to the same RAID group before moving on to the next RAID group? (3 x 256 = 768)
  • Or will striping have a higher priority than FAST VP and write in 256kB chunks, ignoring the 768kB?

How will data actually be written in the pool?

No Events found!

Top