Questions and Answers :
Preferences :
particular download order to avoid wu deadline ???
Message board moderation
Author | Message |
---|---|
TOM Send message Joined: 5 Apr 01 Posts: 53 Credit: 65,422,234 RAC: 86 |
|
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Best advice is to just let BOINC manage the queue. If BOINC thinks there's deadline trouble, it will switch automatically from FIFO to EDF (Earliest Deadline First). It's best not to try to force EDF all the time because this can seriously mess with BOINC's scheduler in trying to balance resource shares across multiple projects over an extended period of time. |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
You should set "Connect Every X" to match the interval that you allow between connections. BOINC will attempt to ensure that every task is completed this far in advance of the report deadline so that it can be reported no later than the last connection before the report deadline. This setting will also attempt to ensure that you have at least that much work on hand. BOINC WIKI |
TOM Send message Joined: 5 Apr 01 Posts: 53 Credit: 65,422,234 RAC: 86 |
|
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
Let's see whats gona happen, there are 50 packets left closed to the deadline out of 4000 in the wu download queue. Are there any messages when Boinc goes to EDF Mode ? The manager indicates "High Priority" for each task that is running that way. BOINC WIKI |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Let's see whats gona happen, there are 50 packets left closed to the deadline out of 4000 in the wu download queue. Are there any messages when Boinc goes to EDF Mode ? You try to feed an Elephant using a (64 kbit/s) teaspoon ;) Can't you get faster Internet? I think you use WU cache of ~7 days. Set the WU cache as 7+0 ("Connect every 7 days", "Additional 0 days"), not 0+7 Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
TOM Send message Joined: 5 Apr 01 Posts: 53 Credit: 65,422,234 RAC: 86 |
You try to feed an Elephant using a (64 kbit/s) teaspoon ;) And the Elephant gets angry about that... ;-) There are many short GPU WU's comming down the line right now... I changed to connect every 10 days and Additonal 0 days and we will see what happens. I'm thinking about faster Internet. It will almost triple the connection costs, because we ware far away from the next connection point, more than 8 km. At this moment, the only way for a faster connection is to go by sattelite, and this is not a real flat rate. The WU's close to the deadline finished in time, but the next deadline will be there. I guess there must be some more intelligence in Boinc to download the WU's witch are closed to the deadline first. |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
You try to feed an Elephant using a (64 kbit/s) teaspoon ;) Set "Connect Every X days" to approximately reflect how frequently you allow BOINC to use the internet (a small amount larger is better than a small amount smaller than reality). Add a few "Extra days" to add a bit of work as a pad. Example: If you allow connections every 3 days, set your connect every X to 3 and add 3 or 4 extra days. BOINC WIKI |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
I think TOM is connected all the time but to very slow Internet connection (comparative to the speed the computer processes tasks). I think in this case it is better to put all the days to the first number ("Connect every 10 days"). This will ensure: 1) You will not receive tasks with deadline of less than 10 days away (so even if tasks stay in download state for days they will not miss the deadline) 2) BOINC will try to get tasks for 10 days WU cache 3) BOINC will try to finish the tasks 10 days before the deadline (using "High priority" mode when needed) (so you will have 10 days for upload and report over the very slow Internet connection) Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
I fail to understand you reasoning on this one. OK, he might want to put part of it into "Connect every X", but not all of it. BOINC WIKI |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Agreed. 2 or 3 days should suffice. |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
You both don't say why you think it is bad to put all the WU cache days to the first number ("Connect every X days"). What problems do you expect? If it is "High priority" mode from time to time this is even desired in this situation (fast computer on slow Internet). "Maintain enough work for an additional X days" did not exist in first BOINC versions and "Connect every" was routinely used for bigger WU cache. Again: what are the disadvantages you see in using 10+0 days WU cache instead of 0+10 or 5+5? (Advantages are in "This will ensure: 1) and 3)" quoted above) But - maybe lower total WU cache (e.g. 5+0 days) will ease the things as fewer tasks will be in download queue (not 4000 as it is now). Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
Potential overload and starvation of long tasks. If the computer always runs in EDF, and is always able to download work, then the following will occur eventually. A long deadline task will be starved of CPU time until it is too late to start the task and complete it by the deadline. BOINC also works harder to ensure that the disconnected interval is filled - even if it means fetching work that will guarantee that all work on the system is reported late or fetching work from projects that have an ever increasingly negative LTD. Using extra days is a hint that the user would like some extra work, but BOINC will not actually fetch work if the situation is already very bad. BOINC WIKI |
©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.