RAID5 SSDs with Areca Card - Mail server

[shockwavecs]

Hey,
I currently have 6 machines that RAID1 with SSDs. They have been running well for about a year. In each system, Drive 1 is an Intel 120GB and Drive 2 is a Kingston SSDNow 120GB. I overprovisioned them to only see 100GB of the 120GB while creating the software RAID (mdadm) in centos. I believe this is the only mechanism to maintain overprovisioning while creating a RAID.

So I now want to build a server that will run Zimbra for ~1000-2000 users.

I have a box built that has 16 bays hooked up to an Areca 1880 card. Originally we beefed it up with a RAID10 across 12 of the 300GB Cheetah SAS 15K drives. 2 x146GB cheetah for RAID1 boot drive. Since building the box, and it sitting for some time, we yanked the 15K drives and migrated them to our SANs.

Now I have a 16 bay server that is pretty much empty and ready for drives.

I would like to create a RAID 5 for this system:
10 SSDs - 5 x Samsung EVO 840 120GB & 5 x Kingston SSDNow 120GB
7+1 Parity with 2 hot spares

It will leave me with 840GB usable space, with 2 hot spares, at a price point of $800. (ignore the price of the RAID card heh)

Overall Questions:
Will this be feasible?
How will the RAID card play with the SSD controllers and mixing drives that use different controllers? (also currently asking benjamin at areca about this)
Any concerns on the size of files that mail servers generate that would eat up space on SSDs? I ask this question based on minimum block sizes, etc.

RAID 5 Questions:
Does RAID5 write to drives evenly like RAID1? Will each drive experience the exact amount of writes as the next drive? (aside from failures and new disks circulated in after a rebuild)

 

[mwroobel]

Even though you are dealing with relatively small disk sizes, I would strongly suggest running RAID6 with 1 hotspare vs RAID5 with 2 hotspares. While I expect you will always of course have current complete backups, the additional uptime it will afford you in case of multiple drive failures is well worth it.

 

[houkouonchi]

Hey,
I currently have 6 machines that RAID1 with SSDs. They have been running well for about a year. In each system, Drive 1 is an Intel 120GB and Drive 2 is a Kingston SSDNow 120GB. I overprovisioned them to only see 100GB of the 120GB while creating the software RAID (mdadm) in centos. I believe this is the only mechanism to maintain overprovisioning while creating a RAID.

So I now want to build a server that will run Zimbra for ~1000-2000 users.

I have a box built that has 16 bays hooked up to an Areca 1880 card. Originally we beefed it up with a RAID10 across 12 of the 300GB Cheetah SAS 15K drives. 2 x146GB cheetah for RAID1 boot drive. Since building the box, and it sitting for some time, we yanked the 15K drives and migrated them to our SANs.

Now I have a 16 bay server that is pretty much empty and ready for drives.

I would like to create a RAID 5 for this system:
10 SSDs - 5 x Samsung EVO 840 120GB & 5 x Kingston SSDNow 120GB
7+1 Parity with 2 hot spares

It will leave me with 840GB usable space, with 2 hot spares, at a price point of $800. (ignore the price of the RAID card heh)

Overall Questions:
Will this be feasible?
How will the RAID card play with the SSD controllers and mixing drives that use different controllers? (also currently asking benjamin at areca about this)
Any concerns on the size of files that mail servers generate that would eat up space on SSDs? I ask this question based on minimum block sizes, etc.

RAID 5 Questions:
Does RAID5 write to drives evenly like RAID1? Will each drive experience the exact amount of writes as the next drive? (aside from failures and new disks circulated in after a rebuild)

1) Should be feasible.... but please run raid6 + 1 hot spare (if you really need the hotspare) rather than raid5 + 2 hotspares... Raid6 is far far more reliable than raid5.

2). That totally depends on your operating system/file-stem/etc... I have no clue of your setup so hard to answer that.

3). In the end the write pattern should be pretty even. Technically when your doing writes which are a partial stripe amount a full stripe of parity has to be written so the stripes that are parity do get written more than the non-parity stripes but parity is distributed over all the disks so in the end this should even itself out pretty good.

 

[shockwavecs]

ok sounds great. I will go the RAID6 route with 1 spare.

Benjamin at Areca said they do not recommend mixing drives.

I'll take that into consideration as well.