NPLB LLRnet server discussion

[mdettweiler]

Originally Posted by em99010pepe

Check my edit...

Okay, cool. I'll make a note about that in the LLRnet FAQ I'm about to post next...

[Mini-Geek]

I've got a few questions: Currently, you've reserved a 320K-333.2K range for the LLRnet server. Are Top 5000 ranges eventually going to be added? Is there going to be any way to choose what range you want to crunch from? Or are you giving up the ability to choose, for the convenience that LLRnet gives?

When running multiple LLRnet instances for multi-cores, is there anything you need to do so that the first folder runs on one core, the second runs on the second core, etc., or does the OS just have them run on the separate cores correctly?

[mdettweiler]

Originally Posted by Mini-Geek

I've got a few questions: Currently, you've reserved a 320K-333.2K range for the LLRnet server. Are Top 5000 ranges eventually going to be added? Is there going to be any way to choose what range you want to crunch from? Or are you giving up the ability to choose, for the convenience that LLRnet gives?

When running multiple LLRnet instances for multi-cores, is there anything you need to do so that the first folder runs on one core, the second runs on the second core, etc., or does the OS just have them run on the separate cores correctly?

For now the LLRnet server will be working on the lower, non-top-5000 ranges, because normally those are less "appealing" to users, and we figured that they would be made more "appealing" by being available through LLRnet. Once we've cleaned up all the non-top-5000 ranges, however, then yes, LLRnet will probably be doing top-5000 work. (I say probably because I don't know what the situation regarding types of available work will be at that time.)

Of course, if you want to do top-5000 work now, you can always use manual LLR. By no means is LLRnet intended to replace the manual reservation system; instead, it is only intended to supplement it.

Yes, the OS will automatically have them run on separate cores. It tries to split up the CPU load as evenly as possible between cores, so no manual monkeying around is necessary. Some users say that manually setting the CPU affinity for each instance of LLR gives performance gains, but it's not proven, and even if it does produce performance gains, they're very small.

[mdettweiler]

All right folks, LLRnet announcement and FAQ have been posted!

BTW Carlos, how long do you think the files that I gave you for the NPLB LLRnet server will last? It was only five files--should I go reserve more?

[em99010pepe]

I don't know. So far 829 k/n pairs were tested in 12 hours. At this pace we still have 7 days to dry out the server but the output has not been constant...I'll go for 4 days...

[mdettweiler]

Originally Posted by em99010pepe

I don't know. So far 829 k/n pairs were tested in 12 hours. At this pace we still have 7 days to dry out the server but the output has not been constant...I'll go for 4 days...

4 days isn't too terribly long, so I went and grabbed a whole boatload of files for LLRnet. I've put them together into one file and attached it here. These, as well as what's already in the server, should keep it plenty busy for a while.

[em99010pepe]

Added...thanks.

Do you want to make some stats?

[mdettweiler]

Originally Posted by em99010pepe

Added...thanks.

Do you want to make some stats?

Okay, sure! I'll go make some stats right now.

[mdettweiler]

Stats have been posted--they're the same as the CRUS stats (I used the same Perl script to make the stats), except that I've added some team stats below the user stats.

Enjoy!

[em99010pepe]

I suppose 321.6 - 321.8 is done from LLRNet side because pairs from 321.8-322.0 (321.880) are already being handed. This means that the server is getting more than 2300 pairs per day...hehe

[gd_barnes]

Originally Posted by em99010pepe

I suppose 321.6 - 321.8 is done from LLRNet side because pairs from 321.8-322.0 (321.880) are already being handed. This means that the server is getting more than 2300 pairs per day...hehe

Just wait until we put it on the top-5000 range! Watch out RPS!!