ESX, Clariion Lun Partitioning Best Practise

[stormvm]

Apologies for the long first post but I require some expert help.

I’m after some SAN and Lun partitioning best practise, our scenario is we have four dell 1955 poweredge blades fibre connected through to an EMC clariion cx3-20 we have recently purchased a new disk enclosure and seven 146Gb drives. The aim is to allocate this storage equally amongst the four esxserver hosts. Previously we have only ever run on local storage.

As regards the virtual machines which will be installed on these hosts they will be application servers not experiencing heavy I/O.

I have read several of the white papers and my initial thoughts were to create a raid 5 group across all 7 spindles giving me a total of 800Gb of usable space in which to create Luns.

The next part is where I could use some help from some of you professionals I was thinking initially to create 4 200Gb luns and present a lun to each of the blades. Two luns controlled by storage processor A and two to storage processor B.

However I would like to ensure that I am using the optimum configuration and also future proofing in that when we require more space which we will. I can expand into that space. To this end I was thinking of creating 16 50Gb luns, then creating a metalun with four 50Gb luns. I.E creating four 200Gb meta luns. Is this a viable option ?

Any thoughts suggestions would be most appreciated

Thanks

[ZMkenzie]

You should avoid putting too many vms on the same scsi queue, so, lun dimension depends on vm size. You should have between 4 and 8 vm per lun for best ratio perfomance/manageability.

If you have a clariion disks from 0 to 4 contain also the storage operative system, so if you create a raid from 0 to 6 you will loose some space (about 40gb per disk) on disks 5 and 6. Anyway, if you have 7 disks a 6+1 array is the only choice.

Hope this helps.

[epping]

Hi

Meta LUNs only offer you increased performance if they go across extra spindels, creating lots of 50GB LUNs and making a meta out of them if they are all in the same raid group will give u no extra performance and is not recommended

the best config for performance on a Clariion is 4+1, u then can create metas across these raid groups, i i was in you position though (i take it that the vault is not located on these drives) i would create one big raid 5 and create a 500GB LUN and a 200 GB LUN, i would go no smaller.

[stormvm]

Thanks for your response guys, the system disks are seperate these new disks are on a different enclosure.

My thought for using metaluns from the offset was to ensure I can expand the metaluns when we require more space. Is this the route to go or if extra space is required eventually on these hosts is it better to create a new lun to the host and create another datastore ?

Thanks

[CXSANGUY]

My thought for using metaluns from the offset was to

ensure I can expand the metaluns when we require more

space. Is this the route to go or if extra space is

required eventually on these hosts is it better to

create a new lun to the host and create another

datastore ?

In my experience MetaLUNs are almost always setup wrong. Even when setup right, they are later altered/expanded in a way that is severly wrong. In fact its probably the most common major mistake I have to cleanup after (and its not often an easy process either).

They are absolutely 100% the best way to maximize performance on a CLARiiON. But you should understand the differences between a Concatenated, Striped, and Hybrid Metalun before you use or configure them yourself. You should also be certain you 100% understand the various way to expand them correctly and incorrectly.

Otherwise you will do yourself more harm then good and would be better off with a simpler straightforward LUN design.

Unfortuneately I've also seen EMC/Dell resources configure them wrong from the outset. Striping two luns together on the same spindles can only hurt performance, not help it.

[stormvm]

>Striping two

luns together on the same spindles can only hurt

performance, not help it.

This statement worries me we have two luns our mail and application servers luns striped across 4 raid groups in a(4+1) (146 Gb disks) configuration. This was done by EMC engineers on initial setup and done as best practise.

There is available space on each of these these raid groups 214Gb which from comments above I was then going to create three 50Gb luns from each of these raid groups to create three metaluns 200Gb and present to three esxserver hosts.

Is this not an advisable way to go ?

Thanks for any advice

[CXSANGUY]

The only way to properly extend a striped MetaLUN is to add another 4+1 set of disks and create equally sized new luns. You should also have and/or be staggering the base lun though in smaller setups thats not too important.

If you wish to extend a striped metalun in any way onto the same spindles you need to alter it into a Hybrid form. So you are actually concantenating the new space as a new stripe and not actually extending the existing stripe. The best way to do this is to choose a component size different than the originals so as to not even have the option of accidentally doing it wrong. That said there are still a plethera of ways to muck up what was probably originally a correct MetaLUN design.

Done correctly this is where a CLARiiON will scream in performance. Done wrong is why many think their CLARiiON is slow and why some piece-o-crap $10 SAN they replace it with is way faster.

PS. Another way, actually a better way often, is to simply create a new metalun stripe of the required size and do a LUN Migration. That will give you an intact striped metalun, not a hybrid, when you are done. Divide the desired final lun size by the number of raid groups and select the base lun by creating the stripe from that point. Also you need to do the math for the metalun properties which a cheat sheet can be found in the pdf guides (or just copy the settings you have now if the spindle counts/raid group #'s are the same)

Message was edited by:

CXSANGUY

[stormvm]

Thanks CXSANGUY,

That makes it clearer for me, I'm still relatively new to this.

For our setup I'm going to create a new raidgroup from the new disks in a 4+1 configuration as per our other four raidgroups and create a lun component in each of these raidgroups sized 150Gb creating a metalun across 25 spindles of size 750Gb and present this to all esxhosts.

[stuart_ling]

To add another point make sure that you have a hot spare or 2 as a disk failure in a raid 5 will impact performance, this does not help the performance of a metalun and can severly impact your ESX server.