slightly ghetto san

[Marcus Lyall]

So here’s where I’m at.

Proavio Chassis with Seagate 1.5 tb drives working, for the moment. ATTO card working.

Procurve switch very difficult to configure but working. On a mac, you need a serial to rj45 cable to do it.
And a degree in computer science. No tech support to speak of. You are on your own.

First Small Tree card was DOA, but replacement now working. Although not in a G5 as hoped.
Technical support from Small Tree very good. Big up Chris.
If in the UK, probably just as easy to buy from Small Tree direct.

So we’re getting an Intel Mac Pro as the server. Not really a ghetto san anymore. My tech friend has had a week of frustration and staring at the terminal window and a large pile of ethernet cables.
Certainly not for the faint-hearted.

Almost there.
Will publish test results when we get the new server. Will also try to put up some details of our process for future reference. Big job on though so might take a while.

[Simon Blackledge]

Post up the commands for the HP 🙂

s

[Marcus Lyall]

OK. So we got the Intel box today, and on a brief test, with fairly empty drives, this is what we got.

3 streams of 1080P Pro-res422HQ playing on 3 machines simultaneously in FCP.

Activity monitor on client machine was showing network in rate of about 90mb/s on each machine. Server showing a peak of 360mb/s network out. This is with a 4 port Small Tree card. Still waiting for the 6 port to arrive.

Very little latency. Maybe a bit less than my local 8 bay connected via the Highpoint. Hmmm.

Will ask my tech friend to post the HP commands, or at least where to find them.

So final config will (hopefully) be:

Apple MacPro single Quad core with 8gb of RAM
Atto R380 raid card.
Small Tree PEG6 ethernet card
Proavio S8MS with 8 x Seagate 1.5tb drives.

Procurve 2810 switch. 6 ports bonded for Server. Other Macs connected via dual ethernet cables. Jumbo frames.

Procurve switch configured with one VLAN for the edit machines, and another for the office machines.
I guess we proved a few things.

1) “Buy cheap, buy twice.”
2) The G5 and Small Tree ethernet cards don’t play well together.
3) You need someone who really understands gig-e switches to sort this out for you. But the rest ain’t so bad.
4) If you do it yourself, allow a week longer than you think you need.
5) It is possible to build a semi-ghetto San with a little help from your friends.

Thanks to Bob, Simon, David, Small Tree and Evan for your invaluable help. I’m sure this won’t be the end of it.

Now, anyone know how to make Bru Server work properly? 😉

[Matt Geier]

Hi Marcus,

Glad to hear things are moving for you for the time being. Let us know how this goes for you, as time will change, so will performance.

BTW – Just out of curiosity, how did you read the latency of the drives, was there a tool you used from the storage?

As for your comment about Small Tree cards and a G5.
It takes about 1Ghz of processor to fully power a single gigabit port full speed. So that means if you want to run 2Gb Ports at 100MB/sec, you need 2Ghz in your server just allocated for gigabit ports. If you can imagine putting a PXG6 in a G5 with Quad 2.44 (a little less than 5 Ghz of processor power) — you could bring it down pretty quick with 7-8 clients connected all going at once… Also compare to the driver now, the driver performance is much faster then it was on the G5. Even your Final Cut tends to process slower on the G5’s then they do now.

(Fun Fact: There are still a lot of Apple G5’s out in the field and lots of them with Small Tree cards since 2004/2005….)

Matt G
Small Tree Forum – https://forums.creativecow.net/smalltree

[Marcus Lyall]

Ah! Didn’t get the whole Ghz per port equation. It all fits!
That would explain the issues we had. Well, anyway, the G5 has been banished to the corner. And the Intel is doing it.

We’ve had some issues with the switch configuration, which have lead to the server disconnecting, but we have a lead on how to sort that out.

We’ve also realised that we need to keep the edit machines online, so we’re going to have to jig things around a bit.

But at this moment, I’ve got 4 simultaneous renders going on in AFX, all pulling files from the server. Which is kinda nice, as long as it doesn’t fall over.

Server is currently spitting out 380mb/s and taking about 12% of the cpu. Rendering to local drives. Sweet. I might even get home before midnight.

[Marcus Lyall]

Quick update.

The ghetto San (not so ghetto now) is still rocking.
It’s currently serving 8 Mac Pros.
Most of them are running after effects, but normally 2-3 are editing HD Pro-res. We are not pushing the bandwidth too much however.

We’ve even packed it into flight cases, shipped it to the US, had British Airways lose it, got it back, plugged it in, and it worked first time.Used it for two weeks. Then shipped it back, plugged it in, and it’s still going strong. Touch wood.

A few issues, here and there. Interestingly, half the machines are running jumbo frames and half not. And it makes no difference despite the warnings not to do this.

The Macs are all configured to run the video server on one ethernet port, and the internet on the other.

Occasionally, you can’t see the video server, in which case, relaunching the finder seems to do the trick.

So all in all, it seems fairly robust. But we have put in a serious backup plan just in case.

[Ian Liuzzi-fedun]

What is your backup plan???

BTW – nice sticking it to Bob