Message boards :
Number crunching :
I Don't Understand (the new wu schedular).
Message board moderation
Author | Message |
---|---|
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
I really do not understand the current wu schedular (and I really do not think I am that thick – but maybe I am). My set up on the particular machine in question is a P4 3.2GHz oc’d to 3.6GHz running 1GB RAM with HT on. Projects and resource share are LHC/50%, Einstein/25% and ProteinPredictor/25%. I was using 4.30 then upgraded to 4.43 just over 24 hrs ago. I have the connect to network preference set at 3 days. Prior to the upgrade I updated all projects for work and obviously LHC has none at the moment. Einstein had 3 or 4 wu’s and PPred had 10 or 15 wu’s. Each project was ticking over nicely and playing fair on actual consumed cpu time and returning work on average 1 to 2 days after I received it. After the upgrade the projects still appeared to get about 50% of the cpu time (remember the projects with actual work are equal on resource share) even though BOINC was continually operating in deadline mode, which I don’t understand since all the wu’s had deadlines that were 6 to 7 days away and remember the reconnect time is at 3 days. Wu’s take approx 1hr on PPred and 11hrs on Einstein. The problem is, I am currently out of work on PPred and Einstein has 3 wu’s, 2 crunching and 1 cached. The current debt situation is; PPred debt = 0 and long term debt = -56700 Einstein debt = 0 and long term debt = -126000 LHC debt = 0 and long term debt = +183000 Questions are; 1. Why was BOINC in deadline mode when the deadlines were 6 to 7 days away? 2. When will BOINC download more PPred work? 3. When BOINC finally does download more PPred work will it only crunch it until it is completed? 4. What happens when LHC comes back up, will BOINC only crunch LHC until it’s long term debt is 0 or negative? Concerns are; 1. Since I’m not always around and due to projects having hiccups I want to keep a cache of 3 or 4 days of actual work, so please don’t tell me to set my cache at 0.5 days or less and things will work as per the old wu scheduling. 2. When LHC comes back I want to crunch the LHC work along side the other project, not try and catch up it’s long term debt. Live long and crunch! Paul (S@H1 8888) And proud of it! |
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
E@H LT debt is now = -216,000 LHC LT debt is now = +309,000 PPred LT debt is now = -93,000 BOINC downloaded 5 Predictor wu's once it ran out of cached wu's. Let's cheer for small mercies, but it does not have even 50% of 3 days of cache downloaded. Here's hoping that a project doesn't go down. Rediculous really. |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
Questions are; Simple, Panic mode is triggered by 2 * connect to setting. PPAH is 7 day project. 2*3=6days, so, one day after the download of PPAH it will be in Panic mode. 2. When will BOINC download more PPred work? It's been my experience that PPAH will download more work even though it has a negative LTD (it's not supposed to) this will cause a higher percentage of PPAH crunched when compared to your desired resource share. 4. What happens when LHC comes back up, will BOINC only crunch LHC until it’s long term debt is 0 or negative? Yup, that's what the current design is supposed to do. Catch you up on LHC even though they are down. Concerns are; Good Luck, I've been addressing this lack of a real cache for some time. the only thing close to a response is when JM7 says"you should set your connect to setting as close to how you really connect" They have never said "NO you can't have a decent cache maintained and still respect resource share" but that is what's happening. tony |
Ian Thompson Send message Joined: 3 Jan 04 Posts: 35 Credit: 182,911 RAC: 0 |
Hi All I am suffering the new scheduller understanding syndrome. I have 2 CPDN WU crunchinging away If I suspend one of the CPDN WU's and Update predictor I get 3 WU they will crunch away happily they don't seem to want to report, unless I click update. The client request more work from the predicator scheduler but I get no work. The log says it has succeeded. This behaviour happens regardless to wether I resume the suspended CPDN WU or not. I can't get new work untill all the predictor WU are completed. I have suspended and resumed a CPDN WU and have clicked update. Also have LHC on this machine what will happen when we get work next week is anybodies guess. Is the new boinc client a part of a conspiracy to throw us off the hiccuping seti scheduler.? Is predictor scheduler different from everybody elses.? If client wants to complete a project work queue. What happens to other project deadlines when the client insists on download multiple CPDN WU.? |
©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.