On Thu, Jan 22, 2009 at 09:17:12AM -0600, Paul Jimenez wrote: > Eugen Leitl wrote: > > Paul Jimenez suggested http://supermicro.com/products/chassis/2U/?chs=826 > > and http://supermicro.com/products/nfo/2_5HDD.cfm which do look > > good indeed. 2.5" is probably more future-proof, provided you > > start small, and then SSD in 2-3 years. > > > > > > My other recommendation is that if you're going to put that much storage > in one box... get two and mirror them. Otherwise you're relying on one > powersupply/motherboard/cpu/etc.
The plan is to use latest FreeNAS builds which support zfs+RAID-Z, and use Unison over SSH over WAN to keep them in sync in diverse geographic locations. Too bad there's no OpenVPN package in the FreeNAS, it would be better if it was a fork of pfSense instead of m0n0. Oh, and it's missing a free pony, too. I'm looking at HD Caviar Green WD10EADS drives to save juice and money (difference is about 78 vs 130 EUR, IIRC), assuming I can use WDTLER to enable TLER on them. > I'm still waiting for NFSv4.1 to get out and stable so it'll be easier I've read the great Clustermonkey article about clusterfs, including NFSv4.1. This is not a cluster application, just an easy >>10 TByte NAS accessed over WAN at about 100 MBit/s speeds or less. Of course it's nice that FreeNAS allows you to use it as an iSCSI target on local GBit LAN. > to build more adhoc clustered storage (which is what you really want > from a redundancy standpoint). In the meantime if you worry about such I'll be playing with PFSVS2 a bit as a Linux vserver storage backend. NFSv4.1 would be an alternative, of course. > things you end up either doing a straight mirror (as above) or looking > into things like Lustre and OCFS2 to provide the SAN-like ability to > have no NAS-head as a bottleneck. And then you end up with NFS or CIFS > 'gateways' so your non-linux clients can access them *anyway*... but at > least if there's OCFS2 or Lustre behind them, it's easy to spread the > load out. I'll be revisiting the cluster fs issue in a year or two. PVFS2 would be great if it also allowed mirrors over nodes, not just stripes. -- Eugen* Leitl <a href="http://leitl.org";>leitl</a> http://leitl.org ______________________________________________________________ ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE
