[K12OSN] New Building's LTSP Server
Terrell Prude' Jr.
microman at cmosnetworks.com
Tue Apr 19 04:47:41 UTC 2011
Joseph Bishay wrote:
> Good day Terrell,
> On Mon, Apr 18, 2011 at 11:42 AM, Terrell Prude' Jr.
> <microman at cmosnetworks.com> wrote:
>> I'd suggest buying the parts and building your own box. You get exactly
>> what you want, it's modern, and you can save some cash over buying a
>> "server" box from, say, HP, Dell, etc. For example, I just built a
>> quad-core 3.2GHz AMD box with 8GB DRAM and a 1.5TB SATA disk drive for just
>> under $400. With careful shopping, you could build, say, a 12-core box,
>> with 16GB DRAM, for about $1200. A single-socket 6-processor machine with
>> 16GB DRAM would run about $600.
> May I ask where you obtained your parts from?
I bought 'em at my local Micro Center on sale. Gotta love those sales. :-)
>> For storage, yes, SCSI drives are very reliable, no doubt. Like you, I've
>> had SCSI disks last 8+ years. However, SATA drives are so cheap by
>> comparison, and disk drive reliability has generally gotten very good over
>> the years. I'd have no big concerns about going SATA nowadays.
> Do you have a preference for hardware vs software SATA Raid?
Yes. I prefer hardware RAID whenever possible. There are two reasons.
The first is that the work of maintaining the RAID falls to the RAID
card's processor instead of your CPU. The second is that the RAID card
will abstract the array so that you don't need to worry about how to do
/boot, for example. It's just easier, especially for the next person
coming in to maintain it.
>> Since everything's Gigabit, personally, I'd recommend two servers, one for
>> the home directories, and another for the storage. Put them all on the same
>> flat network, just for simplicity, with NFS automount (this is very easy to
>> set up). I'm assuming that Adobe Flash (read: CPU hog) is in play. Since
>> all your terminals are P4's and can handle it, you could run Firefox locally
>> and have everything else run from the LTSP server.
> I do want to consider this option -- I just need to make sure it's not
> overly complex since I may not be the only person maintaining it. I
> also do want to make sure I can run firefox locally since that's about
> 90% of the computer usage.
Automount's a snap. I had to learn it for the RHEL cert exams, and I
couldn't believe how easy it was. Why I hadn't been using it for years
before is now beyond me.
>> normally I would suggest Multi-Linking a couple of Gig-E NIC's in the LTSP
>> server. However, the switch needs to support that, and unmanaged switches
>> cannot do that (it's something you need to explicitly configure). So, that
>> means only one Gig-E link to and from your LTSP server for your thin
> I had specified unmanaged switches as they are cheaper and I've had
> good experience with them. IF it was a deal-breaker or a significant
> performance boost I could swing a managed switch but I'd need to
> understand all the pros / cons.
Pros: you have a lot more flexibility with what you can do. VLANs,
MultiLink, broadcast storm protection, etc.
Cons: they cost more.
>> All you're doing here is simply adding on a second internal LTSP subnet. In
>> your scenario, you could add not just an "eth2", but also an "eth3" and have
>> 22 computers per thin-client segment. That would most certainly solve any
>> bandwidth issues. Obviously one could take this to an "eth4", "eth5", etc.
>> if one wanted to do that and had that many PCI slots. At that point,
>> though, I'd just go for a 10Gb NIC. :-)
> I had played around with this a few years back before I learned that
> my switch can't handle it. A point that was raised at the time that
> the PCI bus on the server wouldn't really be able to suppose two
> gigabit NICs at full-speed anyways so it's not that relevant. Is that
> still the case now or is it a matter of the type of motherboard being
> used, etc?
Depends on your mobo. Back in the dual Athlon MP days, I had a Tyan
Tiger MP which had little difficulty keeping up with two Gig-E NIC's.
Of course, I was running said NIC's in 64-bit PCI-X slots, each of which
was on a separate PCI-X bus. :-) Nowadays, with PCI-e, bandwidth
concerns are a total non-issue. Today's monster CPU's have little
difficulty pushing multiple 10Gig-E, cards, so multiple Gig-E NIC's, as
I'm describing, should be no problem. The key is to keep them on
separate PCI buses (PCI-e does this naturally, like SATA does).
More information about the K12OSN