Dell PERC6i RAID 10 limit?

[Asteroza]

This is probably a dumb question but,

Dell's online store configurator doesn't allow PERC6i RAID 10 configurations over 3.6TB. Sniffing around on google, it seems this is an artificial store limit (Dell system images don't go over that size?!?), and the underlying PERC6i does not have such a limitation.

Just to check, is anybody currently running an ESXi 4 host with a Dell PERC6i with a RAID 10 volume greater than 3.6TB?

I realize I can bypass the store limit by selecting for RAID5 and just recreating the RAID volume after the server arrives.

[weinstein5]

Probably not since the maximum volume size ESX can recognize is just under 2 TB -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

[J1mbo · Accepted solution]

I'm not aware of any volume size limitation with this controller. However as mentioned above for ESX(i) the array must present LUNs no larger than (2TB - 512 bytes).

The annoying limitation of the Perc 5i and 6i controllers is that for RAID-10 and RAID-50, only a single LUN occupying the entire available space can be created. Hence to use all disks in a single array either RAID 5 or RAID 6 are required, and with SATA drives because of the size-to-reliability issue, that really then limits the choice to RAID-6 - along with it's 6 IO write penalty.

HTH

http://blog.peacon.co.uk

Please award points to any useful answer.

[Asteroza]

I'm not aware of any volume size limitation with this controller. However as mentioned above for ESX(i) the array must present LUNs no larger than (2TB - 512 bytes).

The annoying limitation of the Perc 5i and 6i controllers is that for RAID-10 and RAID-50, only a single LUN occupying the entire available space can be created. Hence to use all disks in a single array either RAID 5 or RAID 6 are required, and with SATA drives because of the size-to-reliability issue, that really then limits the choice to RAID-6 - along with it's 6 IO write penalty.

HTH

Please award points to any useful answer.

I would be correct to assume that your description of using RAID 5 or RAID 6 would still force me to split the RAID array into (2TB-512bytes) virtual disks on the controller to remain compatible with the ESXi max volume size, then either create new datastores or add extents to the installer created datastore1 using the remaining free RAID virtual disks, right?

I had been contemplating a Dell R510 with the 8 drive chassis and a PERC6i, filled with 2TB SATA disks in RAID 10 to get decent performance and a little safety, but it looks like that won't happen. Not exactly what I would call a nice situation. Unless I use some CLI commands to do RDM on the raw disks (which isn't recommended) and use an iSCSI VM (such as NexentaStor) to feed the disk array back to the host, I basically have no real hope for doing high capacity moderately high performance storage using 2TB SATA disks as an all local storage configuration. Splitting the array into 4 RAID 1 virtual disks gives me safety but limits me to the performance of a single disk probably since I lose the striping advantageg , even if I have a datastore with extents spanning multiple RAID virtual drives.

I suppose I could try to do something sleazy. Have a USB disk for the ESX root partition, have the PERC6i do a single RAID1 with two physical disks for the primary datastore, then have the PERC6i configured to have 6 one physical disk RAID 0 virtual disks, configure datastores on each, stick a single preallocated VMDK on each datastore located on a single RAID 0 virtual disk, then combine the VMDK's in a VM and do disk array management there (virtual RAID or a virtual ZFS RAID for example), then feed the result of that back to the host via iSCSI. But that's an ugly hack that forces data to travel multiple times and abstracts away the disks from the iSCSI VM so storage management doesn't work well.

I guess I should give up and commit to getting a separate iSCSI server to meet my storage needs? As a general rule, ESXi servers needing a datastore larger than 2TB should consider network rather than local storage instead?

[J1mbo]

You are correct, it is that with RAID-5 or 6 you have the ability to create more than one LUN on the array and hence present LUNs meeting the size limitation in ESXi.

As it happens I'm also strugling with this situation with the perc-6. If your workloads are mostly read, raid-6 will perform ok (like four disks) but obviously random write will be poor (like one disk). Running 2TB mirrors then using datastore extents will give you no control over VM placement. On the other hand, running four 2TB datastores would. But each datastore would deliver read performance of 2 disks and write performance of 1.

If think the 'hack' method could be simplified - boot from USB, run 4x mirrors, 4 datastores, a VMDK on each presented to a VM, software RAID-0 across those, then present as either iSCSI (again noting 2TB LUN limit) or NFS back to the host. It will work, but it is very far from optimal.

Have a look at Dell's other RAID controllers. All you need is one that can present multiple LUNs on a RAID-10 volume.

HTH

http://blog.peacon.co.uk

Please award points to any useful answer.

[J1mbo]

It seems the H700 controller supports what we need:

http://support.dell.com/support/edocs/storage/Storlink/H700H800/en/UG/PDF/H700H800.pdf (pg.87....)

7 When creating a spanned virtual disk (RAID 10, 50 or 60), enter the number of physical disks per span in the PD per Span field and press to move the cursor to the Basic Settings box.

12 Set the virtual disk size in the VD Size field.

The virtual disk size is displayed in GB format.

NOTE: You can use part of the available disk space to create one virtual disk and then use the rest of the disk space to create another virtual disk or disks.

Maybe that helps.

http://blog.peacon.co.uk

Please award points to any useful answer.