Message boards : Number crunching : Interested User with No Understanding
Author | Message |
---|---|
Jonathan Evans Send message Joined: 29 Sep 05 Posts: 2 Credit: 58,220 RAC: 0 |
So probably the weirdest situation one could find themselves in is donating something of importance to a group doing great work and not really having a full grasp of what they're doing with that important thing. But since I know it's a good thing I don't care, I want the R@H folks to use my comp to its fullest potential. I don't know a lot about settings for my computer though. I have the second line of Mac G5's and run dual 2Ghz processors with 1.5Gb RAM, have just shy of 200Gb of available memory, and have no clue how to optimize this great program on my machine. I'm running the latest OSX and am connected to a University LAN. If anyone can offer some advice as to how best to optimize performance for this project, I would greatly appreciate it. Thanks, Jonathan [Edit] - just to clarify, the 'second line' is meant to distinguish between the non-water cooled generation before my G5 and the dual-dual that came after - [/Edit] |
Tern Send message Joined: 25 Oct 05 Posts: 576 Credit: 4,695,359 RAC: 10 |
I used a Dual just like yours at last job... drool. All I've got now are the home Macs, a Mini and a couple of iBooks, plus a <blegh!> PC. The important preferences are "use 2 CPUs on multiprocessor computers" and "leave applications in memory when preempted = yes". If Rosetta is your only project, cache size isn't _that_ important, but I would leave it "low", as little as the default 0.1, or as high as 3 or 4, but definitely not 7 to 10. You may be happiest somewhere in the middle, maybe 0.5 to 1.0 - enough work in cache to keep both CPUs busy all the time, but not enough to cause problems. If you aren't already attached to another project, you might consider that, just in case Rosetta is ever down or runs out of work for a while. You can set their resource share quite low, we won't mind... :-) The one I would recommend for a G5 is Einstein. Their app is Altivec-optimized and the G5's fly on it. Any other questions, feel free to ask! |
Jonathan Evans Send message Joined: 29 Sep 05 Posts: 2 Credit: 58,220 RAC: 0 |
I used a Dual just like yours at last job... drool. All I've got now are the home Macs, a Mini and a couple of iBooks, plus a <blegh!> PC. Thanks for the advice, esp about joining up with another project. Just out of curiosity, since I am a novice when it comes to understanding computer but am interested in learning more, can you (or anyone) explain cache and also resource share. What is cache and what is the difference between .1 or 3? [Edit] How do I set cache size? [/Edit] How would I set the resource share, what will that do? Thanks |
Tern Send message Joined: 25 Oct 05 Posts: 576 Credit: 4,695,359 RAC: 10 |
Hm... I don't mind answering, it's actually quieted down over at SETI at the moment - but I'll give the "short version" - the Wiki goes into great detail on all of this, and just about any other question related to BOINC. ("Teach a man to fish...") The cache size is set on the General Preferences web page where it says "connect to network every x days". This controls how much work you download onto your host, and indirectly how often results are reported (but not uploaded). The default is 0.1, or 2.4 hours, or about 1 WU from each project. The maximum is 10 days, which is ridiculous and if you have multiple projects, can make it hard for the scheduler to meet the deadlines. In SETI Classic, there were 3rd-party utilities to give you a local cache, so if SETI was down, you could keep working. Now it's just built in. Resource share (set on the Project Preferences page) allows you to give "x%" to each project you have attached to a computer. Default is 100, but you can set it to anything, and it's additive, they don't have to total 100. If you have 2 projects and each are at 100, each gets 50%. Raise one to 300, so they're 300:100 and it's 75:25. The time is shared, switching between projects every "x" minutes (also a setting you can control). The screensaver (if any) switches with the projects. On a Dual, if you have two projects with equal resource shares and a very _low_ cache size, it will almost always be working on one result for one project and another for the other. With a larger cache size, it may be doing two Rosetta results or two Einstein results or whatever simultaneously, but "in the long run" your resource share, whatever it is, will be honored. I have no idea what the screensavers will do (you can send me a Dual and I'll be happy to check!), and of course Rosetta doesn't have one for the Mac - yet. |
Paul D. Buck Send message Joined: 17 Sep 05 Posts: 815 Credit: 1,812,737 RAC: 0 |
I simplified my life in one sense. I just use percentage numbers as shares. If I want 100%, that is the chare number for the one project. If I have 2 projects, one could be 70, the other then can only be 30. When you get 4-5 (or more) projects it does help a little bit. Although, since not all projects run on all computers, this is not perfect, it is the closest I have come to it. |
John McLeod VII Send message Joined: 17 Sep 05 Posts: 108 Credit: 195,137 RAC: 0 |
I simplified my life in one sense. I just use percentage numbers as shares. If I want 100%, that is the chare number for the one project. If I have 2 projects, one could be 70, the other then can only be 30. This only helps if you have computers all with the same mix of projects. I have some computers that cannot run one project or another for some reason. (I have a couple that can only run 2 specific projects because of memory requirements). With 11 computers and currently 24 projects, assigning things as percentages becomes impossible. BOINC WIKI |
Paul D. Buck Send message Joined: 17 Sep 05 Posts: 815 Credit: 1,812,737 RAC: 0 |
This only helps if you have computers all with the same mix of projects. I have some computers that cannot run one project or another for some reason. (I have a couple that can only run 2 specific projects because of memory requirements). With 11 computers and currently 24 projects, assigning things as percentages becomes impossible. Well, as an absolute you are correct. In other words, I cannot say abostively that I am doing only 1% of all my computing resources on project "x" when I set it to 1% ... I have the same situation where I run different mixes on the machines. Though I only have two basic set-ups. And today I am "fiddling" with the settings again ... The PowerMac is going to do EAH and SAH at 80/20 because theprojects have supurb optimized applications for the Altivec accellerator. On the Windows XP machines it is going to look like: 20% CPDN 05% SAH 40% EAH 01% PPAH 05% LHC 20% RAH 01% SDG 01% PG 07% WCG And for the project team, the 1% projects got there because of the point of the project is low impact and because of the communication thing ... but, for the lack of interaction with participant questions PPAH would be in double digits (for example). The have issues with some work unit types but are not talking to the participants who are asking questions ... (just to drive the point home). ==== edit Ooops ... SO, you can see the that big stick is EAH, though that will likely change when RAH gets an Altivec enabled application and/or starts to run actual work ... I want to go bug hunting ... |
Tern Send message Joined: 25 Oct 05 Posts: 576 Credit: 4,695,359 RAC: 10 |
PPAH would be in double digits (for example). The have issues with some work unit types but are not talking to the participants who are asking questions ... (just to drive the point home). What is driving me nuts about Predictor, is they ARE present in the forums - answering the _simple_, _standard_ things, but ignoring any thread that points out there are some serious problems in their application and work units right now. The volunteers can handle the "why isn't SETI sending me work" stuff that they ARE answering, but "you sent me a result that running 24/7 could not meet deadline, finished 15 minutes past after umpteen hours and got 0 credit", they're totally ignoring. I have them on no-new-work and when current result finished, suspend. If they haven't answered some of these by this evening, they join SZTAKI in the "detached" pile. That leaves me with at least one old slow Mac in need of some super-short-result project. Probably put SETI on it for now. Sigh. I don't have Paul's farm, I can't leave projects around that I'm not going to give at least 50% of a slow machine or 20% of a fast one. |
nasher Send message Joined: 5 Nov 05 Posts: 98 Credit: 618,288 RAC: 0 |
Myself i have multiple projects attached to each computer yes curently i have some of them set to a share of 1% also but that is not cause i dont want to crunch them or such... I am away from my computers for anywere from 1 day to 6 months at a time... with how Boinc works i can remotly set my user share on them ... if there is a problem with a project (say seti.. instead of detaching and reattaching a later date i set it to a lower prioity.. also cause i didnt want to delet the work i completed) with the ability i normaly set projects to try to complete at least 1 work unit per week per project... then if there is a reason i want to do more or less for a project i can remotly on the project site itself reset up the work share and next time my computers connect to them they get the new info.. it is a nice thing fo me well time to get on watch.. hopefully i will be able to be home with my computers in 3 more days |
Message boards :
Number crunching :
Interested User with No Understanding
©2024 University of Washington
https://www.bakerlab.org