May DC Commandos

S

SazanEyes

[H]ard|DCer of the Month - January 2011
Joined
Jan 14, 2009
Messages
1,250
I believe the BOINC project for this month is NFS@Home. We had a good push starting at the end of last month, and it would be nice to continue it. We've gained 10 spots in the last month to move up to 21 in the project, but we should be able to gain at least 7 more this month.

Note that this project can use a lot of RAM. Here's some information from Razor_FX_II on how to control that:

In your NFS account preferences [on the project web site] you can select work units with varying memory requirements.
There may not be work available for some work unit types.

Run only the selected applications:
- lasieved - rarely used so work is rarely available, only for small numbers: (yes/no)
- lasievee - work nearly always available, uses up to 0.5 GB memory: (yes/no)
- lasievef - occasionally used for huge factorizations, uses up to 1 GB memory: (yes/no)
Click to expand...

If you're not interested in math projects, there are two physical science projects were we could also use some help. AQUA@home is a BOINC project, and Muon1 DPAD is a non-BOINC project where we're making steady progress. Thanks to both Briliu and Mr. Pedantic for their big recent contributions to Muon1.
 
Thanks for the nod sazan :) I personally find the muon1 project fascinating so I'm a little partial to it (hence the continued crunching! Plus my workstation becomes unusable for what i need it for with the more intensive projects so I am limited in my projects). We're about to gain a spot in muon as well, a long time coming and still 4 days out.

I'd enjoy some competition in PPD though! I know someone can muster up something crazy :)
 
How does all of this work and what kind of hardware makes a real difference as far as contributing?
 
I cannot speak to the other projects that Sazan referred to but as far as muon goes you head over to the muon1 website and download the latest client (v4.45 as i write this) zip file.

Extract the zip file to a folder where you want muon1 to reside, create a text file named user.txt.

In that file put your desired user name with the team prefix in front of it save that file when you're done. Mine for example is:
[H]Briliu
Click to expand...

Next open up the config.txt file (or config_net.bat or config_nonet.bat for a command line approach) and ensure the settings are to your liking (you would put proxy details in here if required). By default it is set to run using all available threads at 100% under low priority. Save and close that file when you're done.

Finally, you can run a few different files to actually start crunching numbers for Muon1.
  • Muon1.exe - Opens the visual client where you can see the particles progressing through the accelerator (VERY cool, imo, but is full screen and prevents you from doing stuff on the computer).
  • muon1_background.exe - Loads the program as a service in the background for crunching. Ideal for borging family members computers.
  • muon1_cmdline.bat - Runs Muon1 and opens a command line window for tracking. This is my personal favorite, I know it's running, I know it's not having issues and I can close it easily if I need full computer power for FEA analysis at work.

When you get up to 100k of results or 50 finished datasets it will manually connect to the server and send in your results. Takes about a day per connect on my workstation.

As for performance of hardware my Q6600 is ~65k PPD, the quad i have at work is ~110k PPD and the hex at work is ~200k PPD. More accurately here is a benchmark thread on the website. and the latest (as i write this) benchmark graph:

(kpts on the graph = "Point" when i refer to PPD above. My Q6600 gets about 0.820kpts/s)
20101027.png


I opted for details on startup in case anyone skimming through wants to start up :)
 
muon is something I'd like to play with some day but I'd have to do it via wine on my Linux boxen. I'd also have to run 24 instances on my SR-2, little work to setup. :rolleyes:
 
Why would you have to run 24 instances, Tobit?

I've run it with 1 command line instance on the 12 core (2p hexes) at work and it loaded up all 12 without issue. There has also been testing on the forums for people testing 2 or 3 versions ago with 6 single core instances vs 1 hex core instance, with identical (or statistically insignificant, anyway) results.

Let me find the link for you.
 
Briliu said:
Why would you have to run 24 instances, Tobit?
Click to expand...
I thought I read on the muon1 faq that it wasn't multi-threaded. Guess I was wrong. It's been a long week working on too much, but much needed, FAH stuff. :eek:
 
For the BOINC projects, install the BOINC client software. It's very user-friendly. When installing, be sure to uncheck the screensaver option since you'll probably want to run it all the time. It asks you to reboot when finished installing, and you'll notice that it runs automatically when you start Windows. I always switch to the "Advanced View" of the BOINC client, and from there the Options and Preferences menus (under Advanced) let you turn off auto-start and control resource usage, etc.

Add a project from the Tools menu. You can create a new account for each project when you join, but after it's added it will open the website where you can enter your username and join a team. For NFS@home the team name is [H]ard|OCP but some other projects have variations on the team name.

BTW, I had one machine that wasn't getting any NFS work. I thought it was because I had set it to only get the 0.5GB WU, but it was because I wasn't letting BOINC use enough disk space. This was on a test Windows Home Server box which has a tiny C: partition.
 
Muon1 is multi-threaded, but it doesn't really scale that well. I tend to run it on older hardware with fewer cores like a Q9550. I'm going to bring a second Muon1 box online this weekend after I finish some GIMPS work. You could try it on the SR-2, but I'm guessing it would only load the processors to maybe 75%. BTW, Aqua@home is a multi-threaded BOINC project and it seems to scale well, but I haven't tried it on the SR-2 either.

I started up some Climate Prediction as a test, since I'd never run it before and I hate that our team is in the low 500s. The WUs take a really long time to run, but you get points as they run instead of having to wait for completion. It's a low priority project unless it looks like we can gain a lot of spots quickly.
 
Tobit said:
I thought I read on the muon1 faq that it wasn't multi-threaded. Guess I was wrong. It's been a long week working on too much, but much needed, FAH stuff. :eek:
Click to expand...

I wasn't able to find the thread where they did the testing, but after your post i did remember that (prior to me starting) muon1 was single threaded. In version 4.21a he added multithread and has continued to improve it all the way up to 4.44 was his latest multithreaded optimization. So you were correct, but he's come a long way :)

From the Muon1 FAQ:
You can also change the number of threads the program uses for its calculations: this is for multicore machines and you should set the number of threads to the number of processors you have (or double that number if you have hyperthreading).
Click to expand...

Give it a shot, I'd love to see the numbers out of that SR-2... /drool


Sazan,
I have a question on BOINC and the autostart on windows load. I had an issue where it loaded in the background of another user when they logged in. I wasnt aware of it until i was troubleshooting an FEA solve with him and saw it running at 100% when nothing was solving (he's a co-op that isnt there all the time so i run it on the computer when he isnt there and over night). Was there a setting i missed to prevent it from doing this or is it inherent to multi-user workstations and windows?
 
Maybe I'll switch my Q6600's that are still running F@H to muon1 but there are a bunch of new SMP WUs that came out for FAH I am trying to beta test. Too many projects, not enough electricity.
 
If you check out the benchmark graph it scales much better with GHz than with cores. The grey outlined bars "behind" the colored bars show kpts/s/GHz. It shows the ~4GHz i7s as better than the higher core count AMD server or the hex i7 even.

Numbers from SR-2 would be gross, for sure, but like sazan said, it wouldnt scale very well. Boo :(

I guess that leans me more towards the 2600k if i upgrade lol.
 
Briliu said:
Sazan,
I have a question on BOINC and the autostart on windows load. I had an issue where it loaded in the background of another user when they logged in.
Click to expand...

I know one of the install options is install for "all users" vs. "my user", but I don't know if this determines whether BOINC runs or not or just who can control it. I almost always turn off auto-start because I switch between BOINC and non-BOINC projects a lot, and I'm the only user on my boxen. I'd have to experiment with it.

I know if you dig around, there are some "hidden" options that you can set for the client using a config file. I did that on one machine to hide the GPU from BOINC because I didn't want to get any GPU work, but I don't really remember how I did it. If you search the web you should find more info.
 
Will check it out :) Turning off the autostart fixes the problem where it runs the cruncher. It still loads the tray object though, which i'd rather it not do. Thanks!
 
SazanEyes said:
I know if you dig around, there are some "hidden" options that you can set for the client using a config file. I did that on one machine to hide the GPU from BOINC because I didn't want to get any GPU work, but I don't really remember how I did it. If you search the web you should find more info.
Click to expand...
In cc_config.txt to hide Cuda or ATI GPUs:

<ignore_cuda_dev>0</ignore_cuda_dev>
<ignore_cuda_dev>1</ignore_cuda_dev>

or

<ignore_ati_dev>0</ignore_ati_dev>
<ignore_ati_dev>1</ignore_ati_dev>
 
Briliu said:
Will check it out :) Turning off the autostart fixes the problem where it runs the cruncher. It still loads the tray object though, which i'd rather it not do. Thanks!
Click to expand...

Make sure the BOINC Manager is not in the user's Startup folder. The BOINC Manager is just the client interface and the crunching is separate, as you discovered. Usually unchecking "Run Manager at login" should take care of it, but it might have issues with multiple accounts.
 
Been a way for a bit. (Taking Drag Racing photos :))

Still working on NFS and OGR-27. If anyone else has some free cores and is looking for a project OGR-27 could definitely use the help. Its just me and RIUM+

Also, Jathanis I should be mowing you down soon for overall Boinc points. :D
 
Sorry guys. I haven't been around much either. I started a night class and its taking up more time than I had thought.

On a good note, I should have a shiny new Nvidia 550 on my doorstep tomorrow. :) Although its supposed to be around 90F here the next couple days so i might have to wait until later this week to actually fire it up.

I've got a C2Q on NFS and im splitting time on my i7 between NFS and Spinhenge. I'll probably throw the new GPU on either Primegrid or GPUGRID once I get it.
 
Props to Sazan for dominating in NFS. :D He must be causing brownouts in his neighborhood!

Also wanted to congratulate everyone who worked on Einstein@Home over the past month. Looks like we have come a long way and having 9800+ Vault points in very nice. :)
 
etothe3ric said:
Yeah Jeez Sazan what do you have running of NFS to get that many points?
Click to expand...

SR-2 and two i7 980Xs, although one 980X is split 50/50 with Climate Prediction. NFS@home doesn't seem to be very active project, so I'm hoping we can hold on to the gains we're making fairly easily.

I just upgraded the RAM tonight in the L5640 box from 3GB to 12GB, so I'm probably going to throw it on NFS too. I had to reinstall Win7 to switch to 64-bit, so right now it's installing 90 or so Windows Updates. :rolleyes: This was the box that was running slowly on GIMPS, so I want to compare it to my 920 at some point and see if it's faster now. LinX is definitely faster.

We had a nice overtake in Muon1 a few days ago, but it's going to be 8 months or so until the next one. Patience...
 
Last night, I moved one of my stock clocked quad cores from FAH to OGR-27. This weekend, I will convert another one to NFS and another to muon1.
 
I forgot I had Muon1 installed on my SR-2, so I just fired it up. It jumps around a bit, but it's using about 65-70% of the CPUs (dual hex). The load is much better on my 1090T at around 90%, same as on my C2Q. I'll let it crunch a bit and then switch to something else.
 
Looks like this X3220 @ 2,4 GHz is good for ~17K Gnodes in a 24 hour period on OGR-27. Not bad.

Edit: and we moved up one spot to 114th. :cool:
 
Thats pretty much on par with my i7 920 at 2.66. What is your core throughput for nodes/sec?
 
Ok, I have NFS@Home installed on my L5640 and I'll start it up as soon as this F@H WU is done.
Do I only run one instance for all 12 threads?
 
BOINC will automatically request enough WUs to run on all available threads. If you don't have a ton of RAM, you probably want to change the settings on the NFS@home website to not send 1GB WUs. I tried out the big WUs today and it was asking for 8GB or something crazy (for one WU) so I ended the experiment. :eek:
 
Ah, I'm glad you said that. I just changed it so that I don't get the big ones since I only have 6GB in that box. This is a reason to get another 6GB for it though :)
 
Welcome Tobit and Doozer. There are 5 teams in front of us in NFS that will a little more horsepower we should be able to overtake this month. I noticed that 4 of these teams are producing little or nothing so we should be able to keep these positions.

I just got the GTX TI550 up and running. Fired up Primegrid with it so we'll see what it can do. :) After a little testing on Primegrid maybe I'll head over and see if I cant start kicking out some GPUGRID work. Looks like kenlo is out only current GPUGRID cruncher.
 
etothe3ric said:
Thats pretty much on par with my i7 920 at 2.66. What is your core throughput for nodes/sec?
Click to expand...
Generally around a little over 49 million/sec.

Code: 
[May 12 06:00:53 UTC] OGR-NG #a: Completed 27/22-18-3-17-7-12 (42.05 stats  ...
                      0.00:14:08.62 - [49,555,782 nodes/s]
 
Doozer said:
Ah, I'm glad you said that. I just changed it so that I don't get the big ones since I only have 6GB in that box. This is a reason to get another 6GB for it though :)
Click to expand...

I was trying it on my box that just got upgraded to 12GB, and I was still seeing "waiting for memory" messages. The big WUs are memory hogs. I would only try them if you have a huge amount of RAM and either a limited number of cores/threads or if you're running other BOINC projects simultaneously that require very little memory. 1GB per thread is not enough.
 
Need for speed? I love that game!

keep up the good work commandos
 
We have moved up again in OGR-27 to 112th spot. Hopefully take another position or two tomorrow.
 
AAARGH!!!

I've been having intermittent network problems for a couple weeks, but it's been really bad this week with horribly slow upload speeds. (That was one reason I stopped F@H completely a couple days ago, because bigadv uploads were crushing my network.) Tonight I finally had time to start pulling things apart and hooking my laptop up to various parts of the network to isolate the problem. Turns out it was a bad Cat6 cable between my router and primary switch. :mad: I just replaced the cable and now everything's full speed again. I'm now planning how to destroy the offending cable without scaring my neighbors.

Good job on SoB as well as OGR-27.

I'm still working on a web-based DC Vault analysis tool. Hopefully I'll make some progress this weekend. The first goal is to start capturing the twice-daily updates in a database to that I have some history to work with.
 
SazanEyes said:
I'm still working on a web-based DC Vault analysis tool. Hopefully I'll make some progress this weekend. The first goal is to start capturing the twice-daily updates in a database to that I have some history to work with.
Click to expand...

Thermite may be appropriate for that cable. :p

The tool sounds like a great idea. I would love to see the progression and possible a link when you have it up and running. Keep us posted!

In other news I may be shutting down parts of my Garden due to heat. It is going to become more difficult to try and sleep with the heat from my server and desktop due to summer. Thank goodness I am renting a house next year and can hide that server away and a much larger room for my desktop. I am looking at a Dell Powerconnect 2824 for me and my roommates who are all tech savvy to bind it all together next year anyone have experience with the Powerconnect series?
 
etothe3ric said:
I am looking at a Dell Powerconnect 2824 for me and my roommates who are all tech savvy to bind it all together next year anyone have experience with the Powerconnect series?
Click to expand...
If a passive switch will do the job - may I recommend the Cisco 24 port gigibit switch?
I run 12x Majestic-12 nodes through one into my PfSense router and it works flawlessly with over 1000 connections per second on a single port and I am sure it can do much more than that.
http://www.newegg.com/Product/Product.aspx?Item=N82E16833150081
 
Razor_FX_II said:
If a passive switch will do the job - may I recommend the Cisco 24 port gigibit switch?
I run 12x Majestic-12 nodes through one into my PfSense router and it works flawlessly with over 1000 connections per second on a single port and I am sure it can do much more than that.
http://www.newegg.com/Product/Product.aspx?Item=N82E16833150081
Click to expand...
Thanks for the links those look like nice switches. I liked the Powerconnect because it had some management that was relatively cheap at $269 (same as the cisco with the management). I am looking for the management for trying to setup some seperate VLANs, Link Agg and some QoS.
 
Looks like we emptied the NFS queue of 15e WUs. I believe the 16e WUs are the ones with big memory requirements. Hopefully the queue will be refilled on Monday.

edit: I've temporarily fired up Docking@home on BOINC. Docking seems to have very low memory requirements, so on the boxen with 1GB/thread, it allows the NFS 16e WUs to run. On the boxen with less RAM, I'm still blocking the big NFS WUs.

edit 2: NFS@home is back with plenty of 15e WUs.
 
Last edited:
You must log in or register to reply here.
Back
Top