Message boards :
Number crunching :
EDF
Message board moderation
Author | Message |
---|---|
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
If I'm getting "using earliest-deadline-first scheduling because computer is overcommitted", is that because of "one" project or all projects? What is the trigger to "revert" back to another scheduling method and will 4.45 do that automatically when WUs have been crunched? I'm using "no new work" until I know how this resolves. Thanks. |
eberndl Send message Joined: 12 Oct 01 Posts: 539 Credit: 619,111 RAC: 3 |
EDF happens when the time it takes to crunch a specific WU is less than twice your connect time. For example, with a 0.5 connect time, a Einstein unit with 17 hours left and due in 1 day will go into EDF. The reason it does this is to make sure that even if you aren't always connected, (ie on a modem) you will be able to return the WU before it's time limit expires. And yes, the system will go back to "round robin" or short term debt ordering once either the unit is completed. And you don't have to go into "no new work" because BOINC actually does this for you automatically when you're in EDF mode. |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
I read the WIKI several times and still don't understand what happened. I changed from connect every 3 days to connect every 7 days. The computer downloaded more WUs (all 3 projects I'm running) and then determined after the download that the computer is "over committed". I was under the impression that it shouldn't download what it knows it can't complete. |
Keck_Komputers Send message Joined: 4 Jul 99 Posts: 1575 Credit: 4,152,111 RAC: 1 |
Due to the deadlines at all projects, except CPDN, a 7 day queue will put you into EDF mode and keep you there. Another consideration is that the more projects you attach to the larger your queue will be with the same setting. BOINC WIKI BOINCing since 2002/12/8 |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
Ahhh, that is an understandable explanation. I'll revert to 6 days cache and see how that works out. Thanks. |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
In 4.45, there are 3 causes for EDF. 1) A deadline within 24 hours. 2) A deadline that is less than 2 * the connect rate (connect every x days). 3) For any WU, if the sum of the remaining processing times for all WUs with deadlines earlier than or equal to the deadline of this WU is greater than 80% of the time to the deadline for this WU. Once the situation is resolved by processing the WU(s) in question, the CPU will drop back to round robin. Einstein has a 7 day deadline, and to avoid EDF if you are attached to that project, you will have to have a connect rate less than 3.5 days. BOINC WIKI |
RandyC Send message Joined: 20 Oct 99 Posts: 714 Credit: 1,704,345 RAC: 0 |
EDF happens when the time it takes to crunch a specific WU is less than twice your connect time. For example, with a 0.5 connect time, a Einstein unit with 17 hours left and due in 1 day will go into EDF. I think you've got it slightly wrong. ...EDF happens when the time to DEADLINE is less than twice your connect time. Final Classic total: 11446 WU Classic CPU hours: 72,366 |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
Correct. BOINC WIKI |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
Which projects would allow proper scheduling for the optional 10-day cache? |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
I have a 10 day cache set up for my "work" computer. This is the computer I have back in my "office"(in my house, in South Carolina). I set it that way since I'm in Michigan and my wife hates Boinc, but she uploads my work (dial up) for that puter. I am attached to Seti and PPAH on that puter. A 10 day cache nets me roughly 3 days worth of work. I don't know how to actually get a 10 day "real" cache, unless you want to abandon the new scheduler and revert to say 4.32 or earlier version of Boinc. I see were the Dev team is working on "Patching" the Estimates within the projects for some future version of the new scheduler. The problem they are having is that Einstein underestimates the time a WU takes, but the rest of the projects overestimate the time it takes. Anyway, If you want a 10 day cache that is closer to a 10 day cache, then revert to a Boinc version prior to 4.35. If you want to run the new scheduler then expect a variation in the actual crunch times until this is figured out. If you don't like the "EDF" then run one project and set it to any figure you wish. A 10 day cache with the new scheduler will ALWAYS be in EDF, but since you run only one project than it won't matter. does this help? tony |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
It helps. I should quit referring to it as a 10-day cache, when I know that it is just a term for saying the computer connects to the network every 10 days. If that won't work for multiple projects, why allow it as an option when you are attached to multiple projects? When I backed off the 7-day setting to 3 days, it still goes to EDF for Einstein. I guess that's OK, if I know how many days to set the connection to prevent EDF on any / each project. Now I have 10 Einstein WUs on that computer....still running EDF for now. I'll watch to see when it goes back to "Round Robin". I've gone ahead and started making some of my computers work on only one project, so I can control the number of WUs I process for each project. Some computers I've still got configured to run the three projects that I participate in. Thanks. |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
I'm afraid I don't understand the question. help? |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
I guess the real question is: how many days (max) can we use for the network connect for each project? = |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
General Preferences are Shared by ALL projects. Setting a General Preference on any website will eventually (after contact with that project by the scheduler) be transferred to ALL projects. The max as I see it is 10 days. I think shorter deadlined project delimit it to the length of the deadline or somewhere near it(although I don't know for sure). tony |
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
In 4.45, there are 3 causes for EDF. But in reality as soon as you download a wu when you have a preference of 3.5 days, 1 second later BOINC will go into EDF since that wu will be within the constraint of 2) A deadline that is less than 2 * the connect rate (connect every x days). To ensure BOINC does not operate in EDF mode you need to set your connect time to at least the estimated completion time of 1 wu less than 3.5 days. But this is just IMHO. I have no idea why BOINC would want to limit maximum cache sizes to this (and JM7, please don't say to help dial-up users since my laptop is pseudo dial up and you know all about it). Paul (S@H1 8888) And proud of it! |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
Tony, Don't the "venue" preferences over-ride the "general" preferences? Thanks. |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
Yes, they should. There are Primary Preferences(Default) under General preferences, then there are the Venue preferences. Venue changes won't take effect until the Scheduler successfully contact the server of the project where the changes to venue are made. you also need to change the "host" to the venue setting or it won't work. does this help? tony |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
I think a "change" of venue may require a modification on the specific project URL, but I believe a change in venue preferences can be made on any project URL and I haven't tested to see if an update to the venue preference can be made from only the affected project URL or the project URL where the venue preference change was made. I generally make my preference changes from the BOINC / SETI URL, except for the change in venue that needs to be done on the specific project URL. BTW, I made a connection change to 10 days on a computer running only Predictor and now I have 48 WUs in cache. Thanks. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
Due to the deadlines at all projects, except CPDN, a 7 day queue will put you into EDF mode and keep you there. Another consideration is that the more projects you attach to the larger your queue will be with the same setting. I added a new section to the Work Scheduler with two hints ... maybe there should be more like this fairly plain statement? |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.