I know with the PERC 5/e and 6/e you can chain and expand md1000's to increase storage capacity for a raid volume. But is it possible to configure 2 md1000 arrays in a raid10 as a single unit so the raid 10 array is stripped across both enclosures? Has anyone experimented with this? If so was the performance decent? I am thinking of repurposing two md1000's of which their data bas been migrated to a SAN and both fully populated with 1tb sata disks into a raid10 configuration to use as an openfiler based san for our development and QA virtualization infrastructure and keep those greedy vm's (about 60) off of our production san. I was trying to get an idea of the possibilities before spending time with it.
I know clustering is supported with md1000's attached to an md3000 but hoping something is possible for the md1000's to take advantage of them without the complexity of two different iscsi targets in ESX to spread the load out (Then I require independent fe servers to interface the arrays).
In the PERC6 user guide it says; "The maximum array size is limited by the maximum number of drives per disk group (32), the maximum number of spans per disk group (8), and the size of the physical drives".
So you can create a raid 10 of up to 32 drives. To achieve this would obviously require 3 MD1000's, so this confirms that crossing enclosures is possible.
Here are my thought on this scenario – Is this a supported configuration (spanning a virtual disk across 2 enclosures) yes but there a couple things you need to consider before going down this road .
Would I consider setting up 2 DAS attached storage devices and creating a virtual disk that spanned 2 enclosures?
The answer for me would be no.
If you were ever presented with a crisis situation, where retagging the virtual disk was the only remaining option to possibly save the large virtual disk, it becomes significantly more difficult to recover previous configuration and save the data.
One thing I would seriously consider is to purchase SAS drives instead of the Sata which have much lower spindle speeds on the physical disk.
This makes a huge difference virtual disk I/O.
Another issue - The MD1000 is not a supported shared storage enclosure on ESX/VMware platform.
Here is a link to the compatibility matrix on VMware’s support site.
If you got the funds, I would look to purchase an NX1950 and now convert those MD1000’s into a ISCSI target.
When you purchase the NX1950 it comes with the ISCSI target software so you can configure your ESX hosts with their initiator pointing back to the one ISCSI target which is the NX1950.
Which means you can now VMotion between the 2 ESX hosts.
So know you have storage provided at the block level and not the PERC level.
Here is a video that outlines Dell’s Unified Data Storage model.
As far as the Raid 10 ,I see more and more of my customers wanting to set up Raid 10 levels and is in my opinion is the fast Raid architecture that provides fault tolerance.
Thanks for the suggestions Rick. Since this environment is for our qa and development departments I am trying to be cost concious reuse existing hardware. Since we already have two shelves fully populated with 1tb drives its pretty hard to justify replacing them all with scsi. Although Vmware does not officially support direct attached shelves I have verified in the past that they do work. If direct connecting does not offer the granularity necessary then the option to install openfiler on an r300 or 1950 exists. Likewise we also have an enterprise licenses for Starwinds iSCSI target so I could take that approach vs. the costs associated with purchasing an nx series nas.
To add to that; the MD1000 (connected to a PERC5/6), cannot ever be 'shared' storage for any OS.
Also; all versions of ESX/ESXi (version 4.0 included) do not support virtual disks over 2TB.
So as 32 x 1TB in a raid 5 (really bad performance when using highly random IO like running multiple virtual machines) would yield 31 x ~931GB = ~28861GB. As the PERC can only carve this into 8 virtual disks, and ESX can only do virtual disks up to 2TB (2048GB), you'll be wasting ~12TB.
I'd suggest to go with 4-disk raid 10 virtual disks to offset the poor performance of 7200rpm drives and also stay below 2TB per virtual disk.
I also have a 'spare' MD1000 and would like to attach it to a PE T300 Server running ESX4 with virtual Windows Server 2008 and installed PERC 5E. I haven't been able to find any advice on how to get the DAS recognised by my virtual server and keep running into the 'not supported' or 'create SCSI target' responses.
Have you managed to attach the MD1000 succesfully? If so, would you mind outlining how you made it visible on the ESX/Virtual Server?
Dev Mgr
6 Operator
•
9308 Posts
•
3 Points
566
1
Posted May 21st, 2009 07:00
Raid sets can definitely cross enclosures.
In the PERC6 user guide it says; "The maximum array size is limited by the maximum number of drives per disk group (32), the maximum number of spans per disk group (8), and the size of the physical drives".
So you can create a raid 10 of up to 32 drives. To achieve this would obviously require 3 MD1000's, so this confirms that crossing enclosures is possible.