XP SMP - Run Over Network

[EvilAlchemist]

8 Boxen on network. All running XP + Dual SMP .

Instead of each client (2 x per machine) using that systems hard drive ...

Have all SMP Clients - use a single network hard drive to store SMP client info ..

This would allow other HD to power down in systems .... plus cause less network activity when using Fahmon (Due to program only scanning one HD instead of 8)

Would this work?

If so, is it a good or bad idea? Save power ??

 

[Kendrak]

I have no clue how to do it..... (not that good with linux)

This looks like what you might want.

http://reilly.homeip.net/folding/linux.html

 

[EvilAlchemist]

Linda like that ... but with a twist ..

The Boxen are not going to be network booted.

Only the SMP DIR would be on the network machine.

The system would still load XP from the system's hard drive in a boxen.

 

[alan2308]

I don't see any reason why it wouldn't work. Once set, the drives should spin down after a set amount of inactivity, and without any reason for the box to hit the disk again it should stay down.

The only concern that I would have is that now ALL of your work is on one drive. One point of failure kills all of your work units and stops production on every single box. And you just know the drive would die as soon as you're too busy to check on it for a while.

 

[p_o_s_pc]

I have wanted to do something like that before but i never could..If you get it working would you please PM me to help me out? Sorry i couldn't help you.

 

[capreppy]

That's a very interesting thought.

One thing I've realized about FaHMon is that it doesn't like (at least for me) Samba Shares. I have it pointing to Samba Shares now and when I reboot my box (holds FaHMon and 2 Linux VMs), it doesn't seem to re-establish the links correctly. I don't know that it has ever done it right. I have to edit the location and basically redo the link.

Putting everything on one drive (on the monitoring box) could eliminate all of that headache. This would be especially true if I would have to re-establish the links EVERY DAMN TIME I REBOOT.

It would be nice if it reduced some of the network overhead, but I have FaHMon updating only every 5 minutes so I'm not sure how much traffic that would really reduce.

 

[Kendrak]

Why not network boot...... save ya a bunch of HHD even after you raid the network drive

 

[marty9876]

I've done it with the uni client. On the host machines have a scheduled task set to run after 10 minutes idle, stop running when machine comes out of idle (obviously not doable with the SMP, this was a transparent setup so to speak). Just have the task target \\share\folding_foler1 etc.

Problem I had is the machines has windows delay write errors. Not sure why, used default 15 min check points. The LAN was not busy at all, ~ 20 workstations light users with decent switches. No crazy VLAN to the moon stuff or anything. Router, switch, computers etc.

OC-AMD did this with many computers with no problem. Not sure why I had issues.

All in all, no reason to do this to be honest. The amount of power saves (Assuming actually spun down) would be very low.

 

[EvilAlchemist]

Why not network boot...... save ya a bunch of HHD even after you raid the network drive

1/2 the boxen are wireless .. not sure that would work.

That's a very interesting thought.

It would be nice if it reduced some of the network overhead, but I have FaHMon updating only every 5 minutes so I'm not sure how much traffic that would really reduce.

Fahmon does not cause that much network traffic ... but when it has to poll over 15 clients over the network .. takes a min ... plus if a hard drive is spun down .... it will spin up .. not that mine spin down now anyways ....

 

[alan2308]

1/2 the boxen are wireless .. not sure that would work.

It is possible but its all dependant on what wireless card you're using. The LTSP guys have demoed it with prism-based cards before. Its slow as hell, but it can work.

 

[BillR]

That's a very interesting thought.

One thing I've realized about FaHMon is that it doesn't like (at least for me) Samba Shares. I have it pointing to Samba Shares now and when I reboot my box (holds FaHMon and 2 Linux VMs), it doesn't seem to re-establish the links correctly. I don't know that it has ever done it right. I have to edit the location and basically redo the link.

Putting everything on one drive (on the monitoring box) could eliminate all of that headache. This would be especially true if I would have to re-establish the links EVERY DAMN TIME I REBOOT.

It would be nice if it reduced some of the network overhead, but I have FaHMon updating only every 5 minutes so I'm not sure how much traffic that would really reduce.

You could simply change each machine to a static IP, it keeps the net work more clean as well

 

[Tigerbiten]

If your running the clients inside VM's, you could put all the disk images on one drive.
Then give each VM enough memory so it does not need to disk cache.
But I don't know how much extra load that would put on your network.

Luck ............

 

[leSLIe]

8 Boxen on network. All running XP + Dual SMP .

would u use a VM with those Dual SMP clients on XP?
cuz apparently that's the only way to make the clients work without problems.

Now that i mention it, what would happen if u run one SMP client under one user and then run the other one under a different user? hmm... gonna have to try that

 

[EvilAlchemist]

As it stands right now, I run Dual SMP native on XP.

There are no problems - unless you need to close the clients - which i only do when the power fails ....


Since each install is under 20MB .... the network read should work ....

 

[chrisf6969]

if you're that concerned just make the system use that small Linux distro from a flash drive. All you need is like USB flash drive for each system instead of a harddrive and WinXP.

 

[[OCF]David]

This is easy done via Linux. I've played with NFS in the past and its almost completely transparent that your drive is on the network.