Best Drive Raid and Partitions for 3 Drive Server

[raj]

We have the following hardware configuration
Dell 1750 1U
Dual Xeon 3.06 Ghz
2 GB Ram
3 x 73GB - 15000-RPM SCSI Drives
Hardware Raid

I know RAID 10 is recommend everywhere and RAID 5 is not, but with 3 dives only RAID 10 is not an option and we want stable/fault tolerant system.

So what kind of RAID and OS Partitions can we have for Zimbra OSS server which will be loaded with 300-400 POP3/SMTP users and 20-50 web mail users, no IMAP, calendaring is not used by 97% of the users but is enabled in the system, AV/AS is enabled.

What options of the following are best? Performance is not the top most priority but stability and fault tolerance is.

Raid Options:
Option 1: RAID 5

Option 2:
Disk 1: No RAID and will have OS on it
Disk 2 and Disk 3: RAID 1 with /opt on it to have all Zimbra files

* If RAID 5 is slow in writing how much it will degrade the performance in terms of % compared to RAID 10

OS Partition Options for CentOS 5.x
/
/boot
/var
/var/log
/tmp
/opt

Thanks
Raj

[uxbod]

Well with only three disks I think the safest option would be to go with RAID5. With respect to IO it will depend on how many emails are being processed per day. You could boost inbound performance by putting the Amavis work directory into TMPFS as discussed here :- Amavis RAM Disk Setup - OK Now To Reimplement?. As you are constrained by the chassis for the number of internal disks then why not increase the amount of memory to 4GB as that is fairly cheap. This would mean that ZCS would be able to cache more aswell I believe and you could also tune up some of the Java/MySQL and Postfix parameters to make them perform a bit quicker.

[Klug]

Instead of RAID5, I'd rather do some RAID1 + hotspare.
It's much faster and doesn't use more (or less) space.

If one of the two main drives fails, the hotspare will take its place.
If the hotspare fails (it happens), you just have to change it...

[uxbod]

Faster for writes yes, but would it not also depend on how many emails are being delivered each day ? if a user has a 20GB mailbox but only receives a couple of emails a day (large attachments) would RAID5 now be better ? or even better get a small SAN

[raj]

Hi..Thanks for the replies..
The main reason for asking this question was that on 2 servers in past 1 month we got the following EXT3 file system error on zimbra
EXT3-fs error (device sda3): ext3_journal_start_sb: Detected aborted
journal Remounting filesystem read-only

After this whole OS was mounted as READ ONLY and Zimbra crashed.
We tried to find out if the drives were bad..but after running fsck no serious issue was reported other than some missing inodes and stuff.
These machines did not have any raid but multiple drives.

So if this happens (EXT3 Error for any reason. High I/O or bad drive) on a RAID 5 machine, will OS keep working coz whole idea of RAID is to keep working in case any drive is bad.

Thanks
Raj

[uxbod]

Software or Hardware RAID ?

[raj]

HW RAID - PERC 4di in dell server..i mentioned that in very top

[uxbod]

Doh! apologies

Interesting :-

ext3 fs goes ro after a day or three; nfs sharing issues - LinuxQuestions.org

[raj]

yes i have read almost every post on readonly file system on internet.
everyone says hardware problem..so Q is still the same

imagine if 1 haddrive will go bad out of 3 in Hardware RAID5. Will EXT3 re-mount me to readonly or it will ignore it as Hardware RAID5 is still up on 2 drives
/etc/fstab
LABEL=/opt /opt ext3 defaults,errors=remount-ro,noatime 1 1

errors=remount-ro is i guess default..even if you dont put it, It means re-mount read only on any ext3 file errors

The whole idea with RAID is to protect against drive failue but if EXT3 mounts me RO then what is this RAID good for.
this is the only problem i am trying to solve with RAID, its more of a linux admin q than zimbra


thanks
Raj

[uxbod]

VMware Knowledge Base - View Document

What kernel are you running ?