• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Question about client scaling.

A

Axiomatic

Limp Gawd
Joined
Jun 10, 2004
Messages
452
Hey [H]orde,

Got a question. I am a software engineer by trade but I also do a great deal of work with blade system servers (can't say what company, but think who is #1 right now) and want to know if anyone has set the client up as a scaling application? What I mean is, when my blade systems have free cycles in the late night I would like to see the folding client (more realistic is running the folding client as a service in Server 2003) scale to all the low utilization blades. I've got a lab full of hardware (est. 3000 to 4000 CPUs maybe?) that does literally nothing at night and would love to know if anyone has got the client working in an adaptive scaling infrastructure.

Just for fun this weekend I might just set up a single enclosure of blades and just let it run at 100% cpu for the weekend just to see how much WU it gets done. :D

 
Axiomatic said:
Hey [H]orde,

Got a question. I am a software engineer by trade but I also do a great deal of work with blade system servers (can't say what company, but think who is #1 right now) and want to know if anyone has set the client up as a scaling application? What I mean is, when my blade systems have free cycles in the late night I would like to see the folding client (more realistic is running the folding client as a service in Server 2003) scale to all the low utilization blades. I've got a lab full of hardware (est. 3000 to 4000 CPUs maybe?) that does literally nothing at night and would love to know if anyone has got the client working in an adaptive scaling infrastructure.

Just for fun this weekend I might just set up a single enclosure of blades and just let it run at 100% cpu for the weekend just to see how much WU it gets done. :D

Click to expand...

Did I just read 3k-4k CPUs there? :eek:

Well first, get permission from your boss and if granted :D BORG AWAY!

EDIT: And to answer your question, aince they all use the idle cycles there should be no severe effect at all.

 
First, I'm the boss.. at least of my team and lab. :D
Second, 3000+ CPU's... thats just my SMALL lab.
Third, the only permission I would need it with the IT guys for the bandwidth, and they are all outsourced to India and couldn't find the letters I and T with help from Reader Rabbit. (Not trying to offend people from India, my point is that they are not paid enough to care.) And my colleagues and I have been tunneling with OpenSSH for so long to play games in our RARE spare time and they are totally clueless to our shenanigans.
Fourth, we use the folding client to simulate CPU load sometimes so I'm not sure they would care anyway.

Now I'm not saying I will load 3000+ CPU's as managing that would be a full time job. (Truly.) But realistically I'd like to load a rack or two of bladesystems equaling about 200+ CPU's just to try and get this thing to scale.

Last, my comment on idle cycles meant when the blade servers were not doing their normal tasks of SQL, VMWare, Exchange etc. and those apps are not eating up CPU time I want the folding client to recognize that cycles are available on dynamically powered down blades and I would like to to scale and wake up the other blades and use them too.

Right now, if the bladesystem manager shuts down a blade (for a myriad of reasons, power, bandwidth etc.) I want the folding client to wake the other blades back up and adapt to the unused cycles. Maybe you are right though, maybe I just need to use the scheduler and just turn the blades back on and that should wake up the folding client, but then there goes all my fun coding it to "scale." :confused:

I have an idea already, I could write an applet to dynamically raise and lower the thread priority of the folding client, but I need to know if the folding "service" is adaptable in this manner or my little thread applet will be useless. Changing the priority of a .EXE is easy. Services are a little more tough as I think you use DCOM (and a bunch of other shit) for that and I'll admit I'm not as DCOM savvy as I'd like to be.
 
Wow!

Welcome to the [H]ardest Team in the World!

You have some great ideas. Like Jon855 said... don't do anything to jepordize your job or your relationship with your employer. We always say to get permission... Sometimes there is no one but you who can make the choice as to what is the right thing to do. No one here will encourage you to do anything unethical or immoral.

With that behind me... Fold on! Best of luck getting your blades up and running. That would certainly produce a lot of work for Stanford.



 
holy HELL that would spit out some massive amounts of points....i'd say start testing it on a small scale, then ramp it up as allowed, leaving you to find any issues before you go full scale.

 
Thanks for the encouragement guys. I actually played with the client and the adaptive infrustructure software this morning. This is not going to be easy, but I think it can be done still. Its actually looking like the .exe might be easier to use than the service. I would have thought the exact opposite.

Regardless, I will still set up the client on a blade enclosure and see how many points I accrue over the weekend. Should be interesting. :cool:
 
Axiomatic at EOC

Here is a link to your folding at EOC

I want to watch your points ramp up!

Best wishes on your big project.



 
Wow thanks for the EOC link. Hopefully I can complete at least one bladesystem before the weekend. I've got to say that today I got nothing done on it. Damn work actually got in the way.

I try to run my team like I am an employee for myself as well. Always get the job done but playtime is important even at a team level.

I know that if I work my guys to death, they don't last long, myself included.
 
Axiomatic said:
I try to run my team like I am an employee for myself as well. Always get the job done but playtime is important even at a team level.

I know that if I work my guys to death, they don't last long, myself included.
Click to expand...

Can you be my boss? :D

 
LOL, I laugh at myself when I even use the "boss" term. I am a serious gamer and pc enthusiast and about the most non-compliant person on the planet. But for some reason they made me a supervisor. :)

Anyway, keep you posted on bladesystem 1. Man I really would love to show you guys some pictures of the lab etc. but NDA's are a bitch. I can dance around who I work for and elude to it all day, but its the typing of the actual words that violates the NDA.

OK I'll actually ask my boss if a picture of the lab is alright.

Ohhh I have an idea. Maybe I can do one of those [H] photos of some blades. We actually set up 2 pallets of SAS (serial attached SCSI) hard drives in a [H] once but some guy needed to run a floor jack through the hallway and it started to take a little too much time so we stopped. :eek:
 
Axiomatic said:
LOL, I laugh at myself when I even use the "boss" term. I am a serious gamer and pc enthusiast and about the most non-compliant person on the planet. But for some reason they made me a supervisor. :)

Anyway, keep you posted on bladesystem 1. Man I really would love to show you guys some pictures of the lab etc. but NDA's are a bitch. I can dance around who I work for and elude to it all day, but its the typing of the actual words that violates the NDA.

OK I'll actually ask my boss if a picture of the lab is alright.

Ohhh I have an idea. Maybe I can do one of those [H] photos of some blades. We actually set up 2 pallets of SAS (serial attached SCSI) hard drives in a [H] once but some guy needed to run a floor jack through the hallway and it started to take a little too much time so we stopped. :eek:
Click to expand...

Get the permission buddy... :D Even for photos :eek:

 
Hey guys, I learned a little more about the NDA I'm under. It seems that I can post pics and all that, and can even say who I work for (hp) but I just can't act in an official capacity as a rep from my company in this forum. So don't ask me anything hp related. ;)

So anyway, here is a few pics of our lab and the bottom two enclosures in shots 1 and 4 are the ones I'm setting up. We are moving labs right now so things are a mess but I thought you guys might want to take a look. Normally there would be side doors on all the racks but as I said. We are moving to a new building (and back in to cubes damnit! Argh!!!) and its easier to tear these things down with the sides off.

IMAGE_00001.jpg

IMAGE_00004.jpg

IMAGE_00005.jpg


There is a few new pallets of regular SATA drives in the hallway right now. Maybe I will attempt an [H] shot with those for you guys. :)
 
"Borg them all"... AWESOME!

Well I only got to load up one blade with 3 proc in it before the week ended. So the big score jump wont be until next week sometime. Sorry for the delay guys.
 
4 sockets, but those sockets could be single, dual, or quad core processors. So 1 to 16 cores per blade.
 
Axiomatic said:
4 sockets, but those sockets could be single, dual, or quad core processors. So 1 to 16 cores per blade.
Click to expand...

Oh lord...

16 - SMP Clients
8 - SMP Clients
4 - Regular or SMP Clients
2 - Regular or SMP Clients
1 - Regular Client

That's probably how I would do it of course depending on the chips themselves and the memory limitations there may be.

 
I thinking you guys think I'm in Cupertino, CA but I'm actually in Houston, TX.

The old Netserver teams in Cupertino were canned when HP bought Compaq so the Proliant brand could live on. Proliant was the more successful brand/model of servers anyway. :D
 
Well guys my teams lab move got the best of us. We are tearing things down as I type this. I'll have to get the blade systems up and running after the move. I promise to give you a heads up before I get one going though so we can see how fast one of these things running full bore can eat through Work Units.

See you all in the new year, have a Merry X-Mas and a "hammered" New Years!
 
aldamon said:
I think FLECOM might have some competition soon :)

Welcome!


Click to expand...
Even with all 3000+ processors, he would take quite a while to reach FLECOM who has been folding for years at a high production rate. Mind you, we have the SMP client now, but even so, he might not recruit all the potential processing power available in th lab. If that does happen, Stanford's servers will be very, very busy...

 
APOLLO said:
Even with all 3000+ processors, he would take quite a while to reach FLECOM who has been folding for years at a high production rate. Mind you, we have the SMP client now, but even so, he might not recruit all the potential processing power available in th lab. If that does happen, Stanford's servers will be very, very busy...

Click to expand...

Agreed - He'll need to have well time to catch up but given the fact that supposely all 3k CPUs gets on to folding well let's just say they all earn at an average of 500 points per day per cpu. That's what? 1,500,000 points and yes STANFORD will be busy and also well at this rate it would not even take two months to outfold FLECOM.

Scary nonetheless but it's all for the research so let's fold em 3k CPUs :D

 
Yikes, I mentioned before that keeping the client running on 3000 CPU's would be a full time job. The 3000 CPU's was a off the top of my head number of whats in my lab, not what I plan to Borg. Also, I doubt I will ever scratch FLECOM's WU numbers. Flecoms score is truly stunning and impressive.

I plan to load up a rack (or two eventually) of bladeservers with the Folding client, so realistically maybe 640 to 700 CPU's in the nighttime hours and about 100 in the daytime hours. Sorry but the daytime is for real work and I can't infringe on that.
 
Axiomatic said:
I plan to load up a rack (or two eventually) of bladeservers with the Folding client, so realistically maybe 640 to 700 CPU's in the nighttime hours and about 100 in the daytime hours.
Click to expand...
Well, that's a very impressive number of processors! Even if you don't infringe on the #1 spot, it would come close with that number. Fold on! :)

 
Axiomatic said:
I plan to load up a rack (or two eventually) of bladeservers with the Folding client, so realistically maybe 640 to 700 CPU's in the nighttime hours and about 100 in the daytime hours.
Click to expand...
Well, that's a very impressive number of processors! Even if you don't infringe on the #1 spot, it would come close with that number. Fold on! :)

 
Axiomatic said:
Yikes, I mentioned before that keeping the client running on 3000 CPU's would be a full time job. The 3000 CPU's was a off the top of my head number of whats in my lab, not what I plan to Borg. Also, I doubt I will ever scratch FLECOM's WU numbers. Flecoms score is truly stunning and impressive.

I plan to load up a rack (or two eventually) of bladeservers with the Folding client, so realistically maybe 640 to 700 CPU's in the nighttime hours and about 100 in the daytime hours. Sorry but the daytime is for real work and I can't infringe on that.
Click to expand...

Alright, even with "average of running 300 CPUs" at 300 PPD = 90k PPD you'll be giving us all the run for our rigs although FLECOM's legendary output / ppd / pointownages / and so on probably will withstand your borging attempt at most. Let me be the first one to say this 3k CPU > 300 CPU :) But I understand - work comes first before well borging. Whatever you can do helps by this rule (xCPUAmount > 0) > 0CPU

I get BONER everytime I see your pictures... Err... Wait, I meant the Hardware Porn... Oh hell with it...

Fold On!
 
You must log in or register to reply here.
Back
Top