View Single Post
Old 2008-05-22, 15:57   #12
A Sunny Moo
mdettweiler's Avatar
Aug 2007

3×2,083 Posts

Originally Posted by gd_barnes View Post
OK, we're up to 73 committed cores for the rally now. If you haven't already done so, please put them on port 300 for a stress test later tonight from 10PM-11PM GMT.

I currently have 4 on it and will add the other 14 later today.
I've got one core on the server already, and will add the other sometime before 6PM EDT (i.e. 10PM GMT).

BTW, for rallies I've recently been using a modified LLRnet version that can do batching (i.e. hold a bunch of results in its tosend.txt queue, then send them all in at once when the queue gets down to a specified number). It was actually developed by PrimeGrid for their "Project Staging Area", which is a number of LLRnet servers that they have set up to "test run" some projects that they hope to eventually include in their main BOINC system. (Imitation is the sincerest form of flattery. ) Since one of their servers has very small numbers (n~100K-120K), they developed this version that can do batching in a way not dissimilar to BOINC's method of handling jobs (which, of course, would be quite familiar to PrimeGrid's set of users).

Anyway, long story short, I simply downloaded their batching LLRnet client here, changed their server and port to reflect our port 300 server, and followed the directions on the page I linked to just above to get it all set so that it downloads a queue of 10 workunits, but doesn't grab any more work or send any in until the queue gets down to 5 workunits, at which point it grabs 5 more new workunits and sends in all the cached results. The result: lower load on the server.

Previously, this same functionality could be accomplished with a personal proxy, but that's kinda messy if you only have a few computers. Not to mention that Adam's server freezes up if you connect to it with a personal proxy, anyway.

If any of you want to try this out, by all means do so--its results are completely compatible with normal LLRnet, and it works fine with all LLRnet servers. During rally times like this it may be especially valuable to keep the server load down.


Last fiddled with by mdettweiler on 2008-05-22 at 15:57 Reason: typo
mdettweiler is offline