Pool & Logs

Saurabh,

I think I mis-read your post and you were agreeing with me.

Dinesh,

You are correct that the TDEV (thin LUN) is already striped over the backend.  The main benefit of THICK striped metas is to get the backend IOs over more disks.

The benefit to THIN striped metas is to get more concurrency to and from CACHE, nothing to do with the backend.  You can get up to 2x the cached IOs (reads and writes) with a 8-16 way meta over a single device.

If you don't need a lot of IOs from a single LUN, then you don't need a striped meta for performance.

Sorry Saurabh,

Striped TDEVs can give up to 2x the performance of a concat VP meta, there is no performance impact to striping on top of the thin stripe.

However it is much harder to expand a striped VP meta.

Thanks for information.. One more qustion why every where i can see this note for Thin Stiped Meta "Before configuring striped metadevices, please consult with an EMC performance specialist."

I believe when tht document was published Online expansion of Striped tdev meta was not supported and thats the reason we were only recommending striped meta if future expansion of meta was not required.

regards,

Saurabh

The recommendation in that Exchange document was not for performance reasons.  Exchange 20120 does not have high iops requirements like its predecessors so thin metas can generally give the performance needed.  Given concatenated metas are easier to expand, that's why they were recommended.  There are cases with Exchange 2010, with SRDF/S and offline checksum operations where striped can perform better, so if performance is paramount, striped should always be used. 

I did a little test with Exchange 2010 Jetstress running on ESXi5 which itself runs on a Cisco B200 M2 blade. 34 x 15FC 300G drives were used for that in our VMAX-1 (5875). The .edb database drive is 100GB and I tried different stripe configurations. The key results of Jetstress are listed in the table below.

At the end I decided to go with 8 meta members. I need to create 40 of these devices, how much cache will I block with that?

I have not installed PowerPath /VE yet on the host but I will it soon.

Any other tricks to improve performance?

Cache usage won't change.  There is no cache allocation per volume.

The only difference is the maximum WP that the meta can use.  Each volume can use up to 5% of writable cache.

I was directed to this post from another site where my main question was: why do we still need metas if we have FAST VP. I was thinking that FAST VP would make metas the thing of the past.

simmiy

See my reply in this thread.  You may still need metas with VP or FAST VP.

Useful threads and thanks to all!

Thin pool created from 3+1. Do you have any best practice document for recommended RAID layout for Thin Pools?

I see many useful answers; not sure which one to click correct. clicking all helpful..................

Issue resolved resinstallation of full powerpath and HBA tuning.

When you say "RAID layout" do you mean data devices?  They should have 8 hypers per pool per disk, or the minimum number to use all the capacity of the drive.