Rich Mingin (PLUG) on 21 Aug 2016 13:07:38 -0700

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]


If close together failures are a concern as you say at the end, RAID6 *is* RAID5+hot spare minus rebuild time. The extra I/O is largely offset by the extra disk sharing the load.

On Aug 21, 2016 4:01 PM, "JP Vossen" <> wrote:
Semi OT but my argument is that the OS going on the hardware is Debian...  :-)  Note, Debian Jessie requires "firmware-bnx2_0.43_all.deb" for the NICs on the R710.  I haven't tested the R720 yet.

I want to rebuild my VM server and the hardware I'll probably use is a Dell PE R710 with 6x3T drives and a PERC H700 (2.02-0025) [1].  I can do RAID6 or RAID5+hot-spare, and either way I should get about 11T out of it.  All are 7200RPM SATA disks.

As I understand it, either RAID6 or RAID5+HS can handle losing 2 disks, but:
* RAID6 has a write penalty for the extra parity block
* Hot-spare has a gap while the array rebuilds onto the hot-spare

The use-case is my main VMware server (Debian Jessie + LXDE + VMware Workstation 10.x [2].  The critical VM is my main "services" server (DNS, DHCP, file&print, etc., also Jessie) and the next important one will be my MythTV backend (Mythbuntu 14.04), once I virtualize it [3]. Everything else is just test VMs and whatever.  I do not backup the contents of the MythTV server because it's too big, but everything else is backed up in a few different ways (BackupPC, BoxBackup, rsync).  I've got 1.5T in VMs and 3T in MythTV, so 11T is lots of room.

I don't do a lost of disk-intensive things.  Probably MythTV recording 2 shows at once while playing a third is the worst, but I really don't have a good sense of the relative load and capacities of all the parts involved.

This will be most of my eggs in one basket, so I'd really like tolerance for 2 disks to fail close together.  So...RAID6 or RAID5+HS?  Any other random thoughts?


[1] I might also be able to use an R720 with 8x4T drives, but that has 1 bad drive already and I may end up using that for a $WORK thing.  It was drawing 150-180 watts during a RAID init, but I'd like to re-test with an OS with CPU scaling installed and running.

[2] I know that in theory I could use ESXi, but:
1) I have 100+ VMs with many snapshots, and I don't know of any way to move those VMs from Workstation into ESXi without:
        1.1) doing it manually and
        1.2) losing snapshots (show stopper)
2) The last time I tried using ESXi (5.x IIRC) I found it impossible to do anything useful in it
3) I do not have, nor will I have, any Windows involved in any VM control or management, or anything important

Those last two may have changed since newer ESXi has the web GUI and not the fat Windows client.  But the show-stopping snapshot issue remains.

[3] MythTV is currently on a PE2950 with 6x1T drives in RAID5, but the unit is drawing about 300 watts.  My current R710 with 6x1T RAID5 is drawing about 140 watts, so I really wish I'd bothered to put the kill-a-watt on them sooner.  All have dual power supplies so the test is easy (and yes, I unplugged the non-kill-a-watt supply for the test).

--  -------------------------------------------------------------------
JP Vossen, CISSP | |
Philadelphia Linux Users Group         --
Announcements -
General Discussion  --
Philadelphia Linux Users Group         --
Announcements -
General Discussion  --