LLRnet/PRPnet rally August 12th-19th

[kar_bon]

Originally Posted by gd_barnes

Karsten, was there ever any attempt to update the Windows GUI so that it works with the new LLRnet? I'm sure that would be a large undertaking.

The GUI-implementation was only available for WIN, so because I edited some of the LUA-sources both, Linux and WIN use those same LUA-codes but with different EXE. The "old" EXE uses the LUA so there's no need to update or recompile them.

But the reason why I've omitted the GUI is:
In the old LLRnet-EXE integrated the LLR-test and everytime a test was done, the result was 'sent' to the GUI, too, and therefore updated the GUI-window for several clients running on a box.

The LLRnet2010 do not use the old integrated LLR V3.5-code, but the new cLLR V3.8.1 as standalone program, so the LLRnet client is not aware of any change, which tests were done. Especially if LLRnet reserves more than one candidate to test: cLLR will stop only if all tests are done!

So to manage this, cLLR have to test only one candidate, LLRnet takes the control and update the GUI and back to cLLR testing the next pair!

But this is not doable in the next time.

PS: Another reason it won't doable is LLRnet client with the GUI and cLLR have to run parallel, so the communication is not easy or impossible!

The do-script is only the communication between LLRnet-client and cLLR.
Therefore I implemented text-files to look for primes and results done!

I can write a small script to display all primes found and counting tests done for say 4 clients by executing it when it's needed!

[kar_bon]

Originally Posted by vaughan

What I like about the prehistoric version I'm running is the GUI where I can quickly see the tasks that are crunching on each core, the total list of all cores cached tasks and the completed tasks (with primes at the top of the list {if any}).

Why you want to see which pair is currently done?

The stats listed here is enough for me.
I'm checking from time to time if all clients are running well (-> first hint if not would be a more silent fan).

All tests reserved could be done by a small script as suggested.

The primes are all collected in a separate file!

An idea:

I could extend the do.bat to create a html-page put on the desktop to show such info by doublechclick.

[rogue]

I've give Max some code changes that will log some additional information in the hope of narrowing down the issue. In the log he sent to me there were two client threads that exhibited problems when the issue first appeared. One was sending work to a client and one was connecting. Both of these threads stopped logging at different points. I don't know if one caused the other or if they are tripping over one another (deadlock). A deadlock seems to be unlikely based upon what each of these threads was doing when they stopped logging.

I will be out of town (and unavailable) from Wednesday through Sunday. With luck Max can get a new build running and I can get a log of the server getting "stuck" before I leave.

[mdettweiler]

Originally Posted by rogue

I've give Max some code changes that will log some additional information in the hope of narrowing down the issue. In the log he sent to me there were two client threads that exhibited problems when the issue first appeared. One was sending work to a client and one was connecting. Both of these threads stopped logging at different points. I don't know if one caused the other or if they are tripping over one another (deadlock). A deadlock seems to be unlikely based upon what each of these threads was doing when they stopped logging.

I will be out of town (and unavailable) from Wednesday through Sunday. With luck Max can get a new build running and I can get a log of the server getting "stuck" before I leave.

To all: I've got the code changes and the server is now running with them. Thus, next time we have a "too many connections" issue we should get some valuable debug data on how to fix it.

I've asked Gary to NOT restart the server twice a day as he's been doing until we've nailed down this issue. That's because in order to get our debug data, we need to actually let the "too many connections" issue happen. I'm not entirely convinced that restarting the server periodically does anything to stop this, but so far it's worked and now that we ahve the debug stuff in place, I'd really rather get a "too many connections" issue as soon as possible and get it over with. I hate to do this in the middle of a rally, but this is the only time we have enough traffic on the server to trigger the issue. I'm hoping nobody minds.

When the rally is nearing completion, we'll add up all the downtimes and find out how much total processing time was lost on port 9000. Then, as we did in our last rally, we'll extend the rally by that many hours for port 9000 only to compensate. So it should be fair to all involved in the end.

[vaughan]

Originally Posted by kar_bon

Why you want to see which pair is currently done?

I don't know, maybe because I can? Its just nice to be able to see the four cores loaded with tasks and watch. well maybe for a little while, as they complete and the next pair gets moved from the cache and starts getting processed. Why do we run these projects at all?

Because we can.

[vaughan]

Hey Admins ... where are all the primes hiding? As far as I know only lennart has found one so far in the rally.

Is that what was expected?

Is there a prime drought?

What are the chances of us finding a prime on port 3000 and port 9000 this week.

[kar_bon]

Originally Posted by vaughan

Why do we run these projects at all?

Because we want to find all primes in a given wide range of k- and n-values.

Perhaps these will bring new information of distribution of primes or something else.

And: the DOS-boxes you can watch, too

[Lennart]

Originally Posted by vaughan

Hey Admins ... where are all the primes hiding? As far as I know only lennart has found one so far in the rally.

Is that what was expected?

Is there a prime drought?

What are the chances of us finding a prime on port 3000 and port 9000 this week.

I think 3 is found.

http://www.noprimeleftbehind.net/sta...p?content=port

If you look at the bottom off this page.

Lennart

[gd_barnes]

Originally Posted by vaughan

I don't know, maybe because I can? Its just nice to be able to see the four cores loaded with tasks and watch. well maybe for a little while, as they complete and the next pair gets moved from the cache and starts getting processed. Why do we run these projects at all?

Because we can.

Vaughan, I think what Karsten is getting at is that you can get just as much or more info. from the DOS windows and results.txt files with a 10% increase in speed. We don't want to force you to use something that you don't want to. We just want to make sure you're aware of everything with the new client.

In the DOS windows, you can see the pairs that are currently being processed, what pairs are cached as they are being cached, and even each pair's percentage complete. If you don't like to see each pair's completion percentage (I don't), you can set the # of iterations to something very high. You can also see all pairs that you've processed in the results.txt file within each folder. You actually get more info.

Can someone provide a screen shot of 4 LLRnet DOS windows on a quad-core desktop for Vaughan running the new do.bat client? I think there may be a communication issue here. I'd do it but I can't ever remember how to do screen shots.


Gary

[gd_barnes]

Originally Posted by vaughan

Hey Admins ... where are all the primes hiding? As far as I know only lennart has found one so far in the rally.

Is that what was expected?

Is there a prime drought?

What are the chances of us finding a prime on port 3000 and port 9000 this week.

Yes, we are in somewhat of a prime drought. Although we are behind what is expected, it is not completely abnormal. Keep in mind that we're running a much higher n-range than the rest of the project except for the individual-k drive.

At n=870K on a sieve depth to P=26T, the chance of prime is about 1 in 11,000. A recent check shows that we've processed 80,000 pairs so far (!!) So we should have found about 7 primes by now. Lennart alone should have found ~3 primes so far (actual 1), I should have found ~1.4 (actual 1), and you should have found ~1.1 by now.

Everyone keep sticking with it. We're only 4 days into the 7 day rally. The clump of primes could be just around the corner.

BTW, I had a broken machine for a couple of days that I just got fixed so I should be back up to and avg. of ~170 pairs/hour within the hour. At your current load, I'll be curious to see which one of us does more pairs on Tues.

[vaughan]

Yay! I found one:

549*2^868860-1