New Server, Drive RAID/Partitioning Assistance Please

[quietas]

Yay, I finally got a new server. It's a beast. Dell 2950 III 2x Xeon 5420, 8gb RAM, and 6x SAS 146gb drives. I also get to move up to full Red Hat 5.2 Enterprise 64-bit.

I currently have about 60gb in /opt/zimbra with 250 users. I was thinking 2 drives using RAID 1 for 146gb as / and the remaining 4 drives using RAID 10 at 292gb for /opt/zimbra/store. Realistically that gives me 136gb and 272gb on two volumes.

Now with those 6 drives, is my thinking right with two volumes? If not, what would be the best partitioning scheme?

[Klug]

As I said in another thread asking about the same question, that's not what I'd do (and what I've done several times).

ZCS NE allows you to use HSM, just use it...

Keep a pair of SAS HD for OS + ZCS (binaries, primary store, etc), in RAID1.
Put in a pair of NearLine SAS or SATA (750GB or 1 TB or even more) for HSM, still in RAID1. Or four 500GB in RAID10.
And do your backups to another server (through NFS or iSCSI).

If you need/want to do your backups on the same server, use three pairs of HD : one SAS pair for OS/ZCS, one NearLine/SATA for HSM and the last one for backup. All three pairs in RAID1, obviously.

[quietas]

I've seen the Nearline and that will be an upgrade as space is used. I'm working with what I was given though, it was hard enough to get a new server. Before getting an upgrade the other server had to die in a spectacular manner thus causing all business email to stop completely company wide. Things don't get upgraded here unless major catastropies happen.

What I have to work with is 6x 146 gb drives. HSM to larger drives will have to wait unfortunately. I plan for a storage server (I doubt I can swing a SAN, too $$$) with a tape drive. I'll route backups there later.

[raj]

2 drives using RAID 1 for 146gb

what do you gain with this..OS + zimbra dont use more than 3 GB in total..you dont need more than 2-4 GB for logs or less..
so practically 130 GB or so will be sitiing not getting used

for best performence and redudency..i go with RAID 10 all the way with all 6 drives..you get less storage space in total but all other benifits of RAID 10
by doing RAID 10 for all 6 you get Aprox 400 GB usable and you can use 380 GB (400 GB - 20GB for OS and Zimbra install + logs ) storage very easily.. vs 272GB as per your plan

just my opnion

Raj

[fcash]

That's why I like to advocate putting the OS (/ on Linux, / and /usr on FreeBSD) onto 2x CompactFlash using software RAID1. Then you can use all the drive space for applications and storage, without having to come up with weird RAID array layouts and LVM hullabaloo.

[quietas]

I've seen your compact flash evangelism, but that media is not made for that amount of writes. SSD maybe, but there are concerns over the number of write cycles a CF (or most flash) can do. At minimum temp and log dirs should be located on a hard drive.

Aside from that, compact flash drives are an easy way to get in trouble. It's not standard hardware and would trip up anyone who came after me. Additionally, it would be weird hardware not supported by pretty much any server vendor and isn't offered mainstream. Ideally there would be a pair of 2.5" 40gb drives I could stick the OS on, but that is not an option either.

I had thought of just putting the /opt folcer on an an LVM drive with the 4 drives and the remaining unused spave of the first two, but i am trying to reduce any extra confusion for anyone else who might maintain this system in the next 5 years.

[fcash]

For what amount of writes? You only put / on the CF, you know, the part of the OS that never changes, except during upgrades. It's used to boot the system, and that's it. /opt, /var, /tmp, /home and so on go onto the storage array. That way, you don't "waste" a RAID1-worth of harddrive for the OS.

And it's not "exotic" in any way. They appear to the OS as any other IDE or SATA disk (there are plenty of CF-to-IDE and CF-to-SATA adapters out there). It's not like you put them into a CF card reader. You can even get adapters that plug into PCI slots, making them easily accessible without opening the case (the PCI plug isn't active, it's just used to hold the adapter in the server).

Since they're in a RAID1, replacing one is as simple as "remove old CF, insert new CF, rebuild array". Just like any other harddrive. If you use a SATA adapter, they're even hot-swappable at runtime. At least on FreeBSD, IDE versions are even warm-swappable (you have to do some manual stop/start of the IDE port).

The core OS rarely needs more than 4 GB, usually less than 2 GB. Why "waste" even a 40 GB laptop drive for that? It's very hard to buy small harddrives nowadays, but CF and SD cards are very inexpensive and readily available pretty much everywhere.

We do this with our FreeBSD ZFS servers without any issues. The CF disks appear as normal IDE, and we can dedicate every harddrive in the system to ZFS. No wasted disk space, no wasted drives, everything is dedicated to storage. I'm planning on doing this for our VM systems in the future (why waste a harddrive to boot the OS and start the hypervisor, when all guest storage is on the NAS/SAN anyway?).

[Klug]

Why waste a CF for an hypervisor when you can boot it through PXE ?
8-)

[raj]

i use PXE boot for all my ESXI's ..just use little bit more ram to load the whole stuff but i can use iSCSI targets (so no HD at all in the machine..works good for private cloud) or fully use the local HD's

* CF is not fault tolrent and as explained for the orignal Q..if you go RAID 10 all the way with all 6 drivs your OS and DATA are fault tolrent and your get more place for data..

Raj

[Klug]

That's what VirtualIron has been doing, by design, for years.
It's just... smart!