Hi Team,
Can anyone elaborate a little more what <feeder_query_size> is actually used for. Is there any relation to it and <shmem_work_items>

Cheers
Seth

See http://boinc.berkeley.edu/trac/wiki/ProjectOptions#Jobscheduling for the documentation, if you hadn't already.

---

Thanks Jord. I had looked at that but it doesn't really tell me anything.

For some reason it's implied that <feeder_query_size> should be double <shmem_work_items> but it doesn't explain why.

Is <feeder_query_size> actually used by the feeder? From what I understand the enumeration query is only really used by the transitioner?

Cheers
Seth

I'm not sure why the query size is double the shmem size, I would need to look in the code for that. But the shmem_work_size is what it says the amount of tasks that are held in the shmem array that is shared between feeder and scheduler. The feeders job is it to fill this array with unsent tasks and therefore it regularly queries the database. This is what you limit with feeder_query_size, how much tasks the feeder retrieves from the DB to put into the shmem array. You want it to be bigger than item_size because the feeder needs to make sure that tasks for all enables applications are present in the array.

OK so have determined for a single project/app server the value of both these params just needs to be the same. A larger value for feeder_query_size is pointless.

Cheers
Seth

It does not need to be the same. I wouldn't expect a larger feeder_query_size to be harmful. Even if feeder_query_size is lower than shmem item size it should still work but it would take some more time to fill an empty shmem segment. I'm not sure what you want to optimize for. I guess your feeder is not fast enough to refill the array as the schedulers are faster in sending out tasks. You could try to decrease the feeder sleeping time which seems to be a better value to optimize. I don't know if there is a commandline option for this or if you need to change this in the code.

HI Christian,
Thanks for your notes.

I had experimented before with sleep time, I think adding --sleep_interval N to the feeder command did the job.

With the large fleets of vCPU I'm running I exhaust a shared mem segment of 8000 every 5 seconds so I'm trying to push that limit while at the same time trying to manage my other bottle neck which is the MySQL server. There is no option to increase the hardware specs on the MySQL server, it's maxed out already, so looking at ways to reduce the load on it hence my questions on the size really needed for feeder_query_size.

Cheers
Seth