Message boards :
Number crunching :
Keeping the servers happy
Message board moderation
Author | Message |
---|---|
Dissident Send message Joined: 20 May 99 Posts: 132 Credit: 70,320 RAC: 0 |
I've been trying to reduce the load on the servers from my end and have set my network usage to between 6 and 7 pm daily. I want to go 2 days between connects now, but when I set the 'connect about every x days' to 2 (days), it will still connect every day between 6 and 7 pm. Is this an either/or setting? Have I missed something? |
Alinator Send message Joined: 19 Apr 05 Posts: 4178 Credit: 4,647,982 RAC: 0 |
That's happening because the CI isn't a 'set in stone' kind of value. It means for the CC to assume that it won't have connectivity any more frequently than the value it's set to when considering how much work to DL and when to run it. Alinator |
Heflin Send message Joined: 22 Sep 99 Posts: 81 Credit: 640,242 RAC: 0 |
Does keeping a larger local WU queue (by setting larger 'connect about every' or 'additional work buffer') actually decrease the load on the project servers or help the project servers? Or even limiting when client can talk to server helpful to the project servers? I would think any benefit to the project servers would be very minimal. I’m not concerned about a client running out of work during project down times. Since I want to ensure WU are returned within the deadline, I’ve left my clients settings to defaults of connect every 0.1 days & additional work buffer 0.25 days. So again, is there any significant benifit to the project servers (reduced load, bandwidth, queuries, etc)? SETI@home since 1999 "Set it, and Forget it!" |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Does keeping a larger local WU queue (by setting larger 'connect about every' or 'additional work buffer') actually decrease the load on the project servers or help the project servers? Or even limiting when client can talk to server helpful to the project servers? Yes, though it's difficult to fully quantify. Each time a host contacts the Scheduler there's both network traffic and database queries involved, so to the extent a longer WU queue makes frequent Scheduler contact less likely it is a help. Rom Walton blogged about the database query effect. Each Scheduler contact involves database queries for Host, User, and Team records as well as those needed for reporting results and getting new tasks. A single Scheduler contact which reports 2 results and gets 2 new tasks is about 37% less database load than two separate Scheduler contacts each reporting 1 result and getting 1 new task. Doing 3 at at time reduces the load by about 49% compared to 1 at a time. The more the better, though the greatest savings are only about 70% from reporting 20 or more at a time. For your hosts and selection of projects it's probably not practical to think about more than 3 anyhow. The way that a longer queue affects work fetch (and therefore Scheduler contacts) is primarily through project Duration Correction Factor (DCF). Each time a host finishes a WU, DCF is adjusted and the work fetch calculations estimate time for the other project WUs in queue as longer or shorter. If there are several WUs in queue, an upward adjustment of DCF inhibits work fetch for a longer time, a downward adjustment is more likely to trigger a request for sufficient seconds of work to get more than 1 task. Joe |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
I've been trying to reduce the load on the servers from my end and have set my network usage to between 6 and 7 pm daily. The global preferences override file can contain a <day_prefs> section in which you can set allowed connect times for each day of the week, at least for 5.10.x and later. Using that, you could arrange for 4 connect periods per week at 42 hour start time intervals, or 3 connect periods per week at 56 hour start time intervals. Or you could do 48 hour intervals for Monday, Wednesday, and Friday but have a 72 hour interval on weekends. I'm not sure how that interacts with the overall network usage set via the web or GUI, perhaps someone who has used the feature can advise. Joe |
Dissident Send message Joined: 20 May 99 Posts: 132 Credit: 70,320 RAC: 0 |
The global preferences override file can contain a <day_prefs> section in which you can set allowed connect times for each day of the week, at least for 5.10.x and later. Using that, you could arrange for 4 connect periods per week at 42 hour start time intervals, or 3 connect periods per week at 56 hour start time intervals. Or you could do 48 hour intervals for Monday, Wednesday, and Friday but have a 72 hour interval on weekends.Joe Thanks Joe, I'll give idea that a try. It sounds like the best overall solution. Danke fur alles! |
Dissident Send message Joined: 20 May 99 Posts: 132 Credit: 70,320 RAC: 0 |
Despite setting the daily override, no joy is called. I'm out of ideas. Daily reports will have to do. |
©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.