how much hardware does zfs need?

[PointandClick]

I bought a jetway c7 board with dual gbit and a 1tb wd green for a nas box with the intentions of eventually doing raid 5. Since then I've pretty much been sold on zfs though.

I was looking at a 4850e with 4g of ram. This should be plenty of horsepower shouldn't it? What about for running a vm on top of it? The only native install I've been able to do of opensolaris has been on an old Celeron D so that's not a very good base.

In case I can't get afp working like I want, how is FreeBSD's zfs support these days?

Thanks

 

[unhappy_mage]

I bought a jetway c7 board with dual gbit and a 1tb wd green for a nas box with the intentions of eventually doing raid 5. Since then I've pretty much been sold on zfs though.

I was looking at a 4850e with 4g of ram. This should be plenty of horsepower shouldn't it? What about for running a vm on top of it? The only native install I've been able to do of opensolaris has been on an old Celeron D so that's not a very good base.

That's plenty of CPU power. The important thing performance-wise (and stability-wise!) is that the CPU is 64-bit. 4GB should be plenty of memory. What disk controller(s) are you planning to use? Which board, specifically, are you looking at?

In case I can't get afp working like I want, how is FreeBSD's zfs support these days?

Thanks

Do you specifically need afp? OSX supports NFS and CIFS, which are both available and easy-to-use in OpenSolaris. If you're using an older Mac OS, you can try these instructions for building netatalk, or these, but NFS is so much easier to set up that I'd definitely take it over afp unless you have a compelling reason otherwise.

No idea on FreeBSD's ZFS support, sorry. Last I knew it was on a par with Solaris' implementation.

 

[PointandClick]

I don't have a board picked out yet. By the time I get around to this I may be able to get something better. I haven't looked at controller cards too much. Most boards have have 6+ ports so that should be enough to start with. I was thinking of adding drives in groups of 4. IIRC the less drives you have in a vdev the better?

I haven't tried nfs with OS X, but smb leaves something to be desired. I just assume have the native protocol for my main machine, but I will have cifs/smb going too since it's pretty much universal. I got netatalk/afp going for the most part on a test machine. It still complains about some files, but seems to be working fine.

How does raidz compare to software raid anyway? Everybody around here seems to avoid sw raid like the plague. I don't see how it could be that big of a burden for any modern multi core cpu. Not being hardware dependent seems like a much better solution anyway, at least with zfs being able to get to your data so long as the disks are still alive.

 

[latenorgreat]

NFS seems to be the best option for OSX. It's a bit easier to get up and running than CIFS, and it's a bit faster normally.

Hardware wise just be careful with support, as opensolaris can be picky at times. A 64 bit processor is highly recommended and the more ram the better, 2GB is good, 4GB is better. You really don't need too much horsepower, I'm running a Sempron LE-1150 and CPU usage is minimal pushing files, scrubbing, etc. Depending on what you are using the VM for, the cheapest dual core from AMD/Intel should be fine. The AMD 4xxx dual cores don't have support for frequency scaling, only phenom 1 and older support that feature.

If you go with AMD just check support for the chipset, network, etc. Intel ICH6,7,8,9,10 support AHCI, but I don't think any AMD chipsets support it at the moment.

The FreeBSD support is just ok, I would stick with Opensolaris or Solaris.

 

[unhappy_mage]

I don't have a board picked out yet. By the time I get around to this I may be able to get something better.

Well, consult the HCL to see if a particular chipset is supported.

I haven't looked at controller cards too much. Most boards have have 6+ ports so that should be enough to start with. I was thinking of adding drives in groups of 4. IIRC the less drives you have in a vdev the better?

It's a tradeoff. If you're doing raidz, smaller groups means more IOs per second, but less effective space because you're losing more disks to parity. I'm personally doing 6-disk raidz2 groups, but 4-disk raidz1 groups are reasonable too. I'd tend to go with raidz2 though.

It still complains about some files, but seems to be working fine.

"Complains about" in what sense?

How does raidz compare to software raid anyway? Everybody around here seems to avoid sw raid like the plague. I don't see how it could be that big of a burden for any modern multi core cpu. Not being hardware dependent seems like a much better solution anyway, at least with zfs being able to get to your data so long as the disks are still alive.

I have never seen a bottleneck that was caused by CPU power. However, that doesn't mean I've never seen a bottleneck---make sure you get a motherboard with an x4 or x8 PCI Express slot to put a controller in, because once the onboard ports are used up you'll want a controller that doesn't use PCI.

 

[movax]

I just finished my ZFS server build: Parts
Performance

Might up the CPU voltage a bit, but its still staying under 200W.

 

[PointandClick]

Will definitely be hitting up the HCL before making any purchase.

I'm mostly doing smaller groups because of $. Got to love college. It may be a while before I even go this far, may need to build a desktop workhorse first. No matter how "high end" a laptop is, it's never fast enough.
If I was doing bigger groups I would definitely consider raidz2.

I don't recall what exactly the problem was with netatalk, I'll check it out here in a bit.

Nice info movax. Are those loads typical? What speed do you have it clocked at now?
With my jetway board and single disk I get around 30MB/s, very rarely 40, on big transfers from my Macbook Pro. This is just observing from iStat dashboard. I'm sure there's more there somewhere, laptop drive, cat 5 cable, who knows.

Thanks for all the input.