This post is more than 5 years old

14 Posts

867

July 29th, 2009 12:00

How many Hypers in a Meta Device

What is EMC's Recommended practise for the Number of Hypers in a Meta Device.

Does it matter on the number of hypers as long as care is taken that no 2 hypers overlap over the same physical disk ?

6 Operator

 • 

5.7K Posts

July 30th, 2009 01:00

I think you meant to ask about the number of symdevs in a meta. In a single symdev there are a number of hypers / slices. Each meta consists of a number of symdevs.
I've had discussions with a Symm performance guru about this and what we came up with is that the number isn't that important as long as you don't have an overlap on the same physical disk. With "symdev show " you can check out the physical disks involved with each hyper or slice withing each symdev. There might be another issue about the number of symdevs per meta and I've heard something about 2 to the powers of x.... so 2, 4, 8, 16, that kind of numbers, but I haven't been able to find any white papers about that.

I see you're pretty new to the forums, so hereby I welcome you ! If you have the time, please introduce yourself in the Coffee Break.

:)

6 Operator

 • 

2.8K Posts

July 30th, 2009 07:00

Yes Quincy, thank you. I was talking about "small numbers". But I agree, it's better to warn even on higher limits :-)

6 Operator

 • 

2.8K Posts

July 30th, 2009 07:00

Quincy unfortunatly I don't remember the name of our other thread.. But I remember we talked a lot on this topic (metamember number vs performances vs protection vs number of drives in backend). Thank you :-)

6 Operator

 • 

2.8K Posts

July 30th, 2009 07:00

The reason you found is exactly the reason for pushing toward 2, 4, 8 or 16 members, depending on your backend size :D

However as we already said, it's simply a suggestion. Feel free to choose any metavolume members, as long as you don't reuse same physical drives :-)

1.3K Posts

July 30th, 2009 07:00

Power of twos fit well into the DA/Disk layout.

14 Posts

July 30th, 2009 07:00

Thanks for the explaination.
My understanding too is that as long as there is no overlap on the same physical disk, number should not matter.

But I have heard exactly what you have about the using only 2, 4, 8, 16 members in the Meta and am curious to find where that is coming from. I didn't find anything in Black and White on that rule either.

The only reason that I could think of is that using those numbers maybe provides some sort of balance across the DA's but I do not have any explaination of how it would do that.

6 Operator

 • 

5.7K Posts

July 30th, 2009 07:00

That's exactly what I had in mind.

If your Symm is using all physical disks with about the same load, you should be ok as the DA's and disks balance themselves this way.... You know what I mean. We're on the same wavelength.

1.3K Posts

July 30th, 2009 07:00

I would suggest not making the member count too large, like 250 members might be too large to really be able to get good performance.

6 Operator

 • 

5.7K Posts

July 30th, 2009 07:00

You're welcome :)

14 Posts

July 30th, 2009 07:00

Thank you All for the Responses.

6 Operator

 • 

5.7K Posts

July 30th, 2009 07:00

emcuser123,

are you satisfied with our answers ?

1.3K Posts

July 30th, 2009 07:00

If you made a 250 member meta with 3+1 RAID5 and didn't share any disks, that would be a meta across 1,000 disks. To have 1 IO queued per disk on average, you would need to have 1,000 IOs queued to the meta, and that is assuming a perfect distribution.

14 Posts

July 30th, 2009 07:00

Quincy,

Explaination of that statement would be really helpful.
No Events found!

Top