Need advise for server farm

[mark5365]

I would like to start off with saying thank you for taking the time to read this post, especially if you have some advise for me.
What i have is a 5 poweredge 2950 given to me working with a sick cabinet setup for this project. I want to make sure we can get the most out of them if im going to pay the power bill and upkeep them myself. I think im going to load all 5 with WinServer 2012. I have been folding on my main office server for about a 1 or more years on the old client and just realized that their is a new client out. I have been playing with it on one of the servers i have already loaded. Im looking for some advise on how to do this to save time, i am very busy and need to be fast with deployment. Should i run VMware client? just run 1 client on each server? Can i manage all of this from 1 console when everything is setup?
What is the best, most efective way?

Thank you again
Mark

[PantherX]

Welcome to the F@H Forum mark5365,

If those systems are dedicated, then this is what I would do:
1) Install Linux 64-bit version of your choice (no need to pay extra when a free alternative is available)
2) Setup FAHClient to use all available CPU in each system (clustering doesn't work)
3) On your monitoring system (can be laptop, desktop, etc), you can install FAHControl and then configure the servers such that all clients can be monitored from a single location. If done properly, you can monitor FAHClient across different OS with out any issues.

VMWare client will give less performance than a native client. Moreover, currently, SMP WUs are faster on Linux when compared to Windows. However, this can change with a new FahCore is released (this isn't very frequent).

[bruce]

You said you were going to run WinServer 2012. PantherX recommended Linux. Either one will work. You have to decide if there's a strong reason why you want one or the other, but I would base that choice on whatever else you might want to do with those machines. [If they're to be use ONLY for FAH, then Linux is the right choice, as PantherX suggests.]

[mark5365]

I really want to use Server 2012 so i can play with them while they are Folding. I have very little linux knowledge and would take more time than i have to setup. I will be loading them today and setting up remote access to each.
So i guess i should ask what the best way to fold with them in windows environment would be. I can run single client on them for now.

[bruce]

Running a single client on Windows Server 2012 is a fine alternative. It can use as many threads (CPUs) as you allocate to it and will also fold on recent model GPUs if you choose to install them. On the CPUs, it runs at an extremely low priority which yields resources to whatever else runs on the hardware but will use as much as 100% when other tasks are idle, although the default is to leave one CPU for other tasks. That's probably the only adjustment you MIGHT want to make.

[PantherX]

If you plan to use the servers as a test lab, than do note that SMP will run as fast as the slowest thread. That means that if you have 8 CPUs, and you have set-up SMP:8 plus are running tasks which require about 10% of the CPU, the SMP will be severely effected in a negative way. In this case, it would be better to run as SMP:7 or SMP:6 (depending on how much resources your other tasks would require). Thus, if you could tell us the number of CPUs, we might be able to offer some suggestions to optimize your F@H contributions based on your needs.

Do note that setting up F@H in Ubuntu 12.10 64-bit is a task that can be accomplished completely via the GUI in under 5 minutes. I am not a Linux expert and found it very easy to do so. Having said that, you could have a Linux VM that is running in a Windows environment and still get higher PPD when compared to native Windows but less than native Linux. However, this is more dependent of the hardware in use so YMMV. If lack of time is a concern, you can always set-up in Windows native environment and experiment with VMs whenever you have the chance to do so, if you're still interested.

[mark5365]

What i like about the idea of using VMware is i can pause the VMbox when playing with the servers. I can also save their state when i restart and allocate the power easier. A few years back when i was big into overclocking i found out about F@H i was running a VMware with a pre 4 core setup i got from OCN. i cant remember what the name was, but that would be ideal. Do you guys have any links to VMware save thats already setup for 4-8 cores? I have just installed the Os on the servers and running updates as we speak.

[mark5365]

- Update -

So i only have 3 of the 5 poweredge servers up each with 2 x duel core xeons with 8gb ram running F@H v7.3.6 @100%. Also have my main poweredge 2x Quad xeon running @ 30%. The other 2 units have issues to im going to replace it with a proliant with 4x 6 core optron's with 32gb ram . I have the server just need the c19 power cords ><. Im going to try to get 3 more of thouse in the next few months, ill keep you guys in the loop. I wounder how many points this will give me?

[Zagen30]

Kind of hard to speculate on PPD without knowing what generation chips those are. The 2x2 cores won't be terribly effective. If you run bigadv units on the 4x6 core, which requires you to run Linux, that should put out a couple hundred thousand PPD.

[PantherX]

Just make sure that you are using a passkey (http://folding.stanford.edu/English/FAQ-passkey) so that once qualified, you will be getting quick return bonus points.

[P5-133XL]

If you run bigadv units on the 4x6 core, which requires you to run Linux, that should put out a couple hundred thousand PPD.

Without knowing the specifics of the chips, you do not even know if it will even come close to completing bigadv's in time.

[mark5365]

If you run bigadv units on the 4x6 core, which requires you to run Linux, that should put out a couple hundred thousand PPD.

Without knowing the specifics of the chips, you do not even know if it will even come close to completing bigadv's in time.

What dose that mean?

[mark5365]

Just make sure that you are using a passkey (http://folding.stanford.edu/English/FAQ-passkey) so that once qualified, you will be getting quick return bonus points.

Thank you Panther. I have just setup a passkey, i didnt even know if that was important.

[Zagen30]

Without knowing the specifics of the chips, you do not even know if it will even come close to completing bigadv's in time.

The oldest hexacore Opterons are the from Istanbul line. A number of people over at EVGA have put together those rigs, and they've been able to make the deadlines on 8101 with some of the slower hexacores in the series, no overclocking (don't think anyone's made a custom BIOS for Socket F mobos anyway). There's enough headroom that even the slowest ones should make the deadlines.

I will clarify my PPD estimate, however. On some of the slower hexacores one would see closer to 100-120k PPD on 8101, which I wouldn't classify as "a couple hundred thousand." That estimate would probably be more accurate on the other 810x projects.

[P5-133XL]

If you run bigadv units on the 4x6 core, which requires you to run Linux, that should put out a couple hundred thousand PPD.

Without knowing the specifics of the chips, you do not even know if it will even come close to completing bigadv's in time.

What dose that mean?

There is a special class of SMP WU's called bigadv that give more points than standard SMP WU's. They require very high performance machines with a minimum of 16 cores and have very tight deadlines. Those deadlines are severe enough that many 4 processor machines with enough cores but using older or lower-end processors running 24x7 will still not finish them in time. So without knowing the processors used it is pure speculation as to the points you may get or even the suitability of running bigadv's.