ZFS Setup Questions

R

rst6616

n00b
Joined
Jan 18, 2011
Messages
23
My goal is to build a high performance home server that would be manageable by a newcomer to Unix/Linux. I am not afraid of CLI if part of your recommendation, although using either a built in web interface or napp-it would probably be easiest.

I am trying to decide which O/S to install for my first new ZFS storage system server: I have preferred the following three choices due to the apparent built in compatibility (thru samba) with my other Windows 7 workstations:

NexentaCore Platform3
NexentaStor Community Edition
OpenIndiana

My Server Hardware:
Supermicro X8DT3-F motherboard
-12gig ecc un-buffered Kingston KVR1333D3E9SK3/6G
-Intel Quad Xeon 2ghz E5506 (may add a 2nd cpu later if needed)
-two-320gig (all 7200rpm server grade) drives to be set up for mirrored boot drives connected on the regular sata ports of the motherboard
-five new Sata Hitachi 5K-3000 2TB drives for the primary performance pool attached to the built in LSI 1068E SAS controller

My current data requirements are only 4 TB.

My questions are:

1. Which of these O/S’s would you recommend for a novice and why?
2. Which ZFS raid mode is best for performance?
3. I don’t understand how to label the five storage drives so that when the time comes and a drive inevitably becomes defective, I can easily tell which of the data drives needs replacing. Can someone please explain to me how to do this?

Thanks
 
Nexentastor is maybe a little more user-friendly, but I and others have had stability issues with some features, and no indication they take it seriously. So for nexentacore vs OI with napp-it on top of either, I'd go for OI just because the kernel is newer. I would not mirror the root drives - you should be able to do set up a raid1 root and let zfs do the hard work. For the data drives, raidz.
 
I will wait until Nexentastor has fixed their issues. Thanks for the suggestion. I have just downloaded OI & will try to install sometime this weekend.

If I raid1 the two 320G boot drives instead of mirroring them, does this help performance alot?

Since I have almost 10TB of data drives (I can easily add one or two more 2TB drives if recommended) and less than 4TB of data, I was thinking of setting up 2 vdevs & mirror them instead of just the one.(I hope I am using the correct terminology). If I do this, will this affect my performance and do you think mirroring the two data sets is a good idea?

Thanks for your advice. This is very helpful.
 
Depends on how the HW mirroring works. SW mirroring will be better because zfs can do its juju - and read performance will be better than one drive, since it can spread the reads out. For your data, raid10 doesn't make much sense, since you have an odd number of drives. I suppose you could have 4 drives in a raid10 with a hot spare, but you are losing 3/5 of your storage. A raidz would give you only 1/5 loss with still able to lose one data drive. If the redundancy is a big concern, put one more data drive on and do a raidz2.
 
What I wanted was to use ZFS software raid1 on the two boot drives. For data, if I add the extra 2TB drive and do a raidz2, I understand that this would increase data safety. But would it also affect my performance compared to keeping just one copy of the data with raidz? I plan to do large data transfers often between this storage server and my windows7 workstations. Therefore, I would like to try and keep the network transfers saturated and running as fast as possible. Thanks for your replies.
 
No, AFAIK, raidz2 has no performance hit over raidz1. With those drives in a raidz2, the network should be the limiting factor...
 
My raidz2 is crappy (relatively) 640GB WD blue drives (6-disk raidz2). Reading a large file on the OI box and discarding the data, I can read 3X as fast as the gig network will allow...
 
Use the ZFS mirror for the boot/Os disks. No need to go get hardware raid1 for this. Let the OS do its thing. ZFS boot mirror works great - I've been running this for a couple of months now.

As for performance - the ZFS root mirror will at least equal, and in many cases exceed, performance of a hardware raid. Few but the very highest-end raid cards do any kind of read optimization on raid1 (striped reads). With ZFS mirror you get that improvement for free.

Do understand, however, that performance of your boot/OS drive is almost meaningless for this application. You could run it off a slow/crappy IDE-CF converter and you probably wouldn't notice a difference for the file server application. Once you've booted up, Solaris will run ZFS, CIFS, NFS and most if its kernel from memory and be happy...
 
danswartz: Just to make sure that I understand your advice. You get approx. 1/3 of the max speed of a theoretical gig network speed (eg approx 35MB/s) or 3x max speed of a gig network (eg. approx 300MB/s) when reading large files? And what happens when your reading small files?
PigLover: I am not planning hardware raid for my boot drives, but only software zfs raid1 (zfs mirror). Thanks for letting me know that the speed of my boot drives doesn't really matter. For data, given my existing hardware and my willingness to buy more drives if recommended, what zfs configuration would you recommend if using five, six or more drives in raidz1 or raidz2, given my goal to keep the 1gig network saturated when transferring large amounts of data between the server and window7 workstations frequently? Thanks once again for your replies.
 
No, what I am saying is that my crappy raidz2 is still good enough that the limiting factor on a read across the network is the gigabit enet.
 
You must log in or register to reply here.
Back
Top