PRPnet beta test server

[mdettweiler]

Originally Posted by Brucifer

hmmm..... hourly results for G2000 don't seem to be updating on the "progress" stats page.

I'm putting some systems on the test server. Where are you in the testing phase now? I'm only running the linux client stuff, but I have it on a combination of OpenSuse-64, Mepis-64, and Slackware-32 systems. All systems are multi-core.

Regarding the hourly results: I don't have the full status page set up for the PRPnet server yet, just the daily results file copy-off. In the meantime, though, all the results will still get into the DB, and will show up on the progress report in the correct hourly columns after each daily copy-off.

Regarding where we are in the test phase: currently we are running k=300-400, n=260K-300K doublechek work. It is at about 267K right now with no problems found so far. The plan is to complete this current batch of work and then verify the results with the original sieve file. We're guessing that will probably turn out OK (assuming that no unstable machines are on the server, which there probably aren't), though we'll still check to make sure. After we're done with that, we'll load it up with some very small candidates starting at n=10K; Gary has said that he would be willing, then, to put a bunch of his quads on the server briefly, along with whetever anyone else is willing to contribute, to run a "stress test" and see how much load the server can handle.

Max

[Mini-Geek]

Originally Posted by mdettweiler

I am pleased to announce that results from the G2000 PRPnet server are now being included in our stats database at http://www.noprimeleftbehind.net. Just like with an LLRnet server, you can now see G2000 on the Progress Report table, and the results count towards your stats totals.

Cool, cool. I'm considering using a private PRPnet server in the future, run on my machine, mainly to make it so I don't have to manage balancing cores and can more easily choose just one file, if I choose that. Would that be any problem as far as results processing or anything goes?

Originally Posted by mdettweiler

Meanwhile, PRPnet v2.1.5 was released a few days ago, and though I haven't gotten the chance to update the server from 2.1.4, I'll be doing that sometime soon. I will also be posting client binaries for 2.1.5, as some new features were added to the client, including the ability to set a default startup/shutdown behavior which overrides the respective menus.

Where was 2.1.5 released, or was that a mistake? 2.1.4 is the most recent I'm aware of.

With the way the stats are pulled into the NPLB database, will the bug I described here take effect (i.e. give full credit for abandoned tests)?

[rogue]

Originally Posted by Mini-Geek

Cool, cool. I'm considering using a private PRPnet server in the future, run on my machine, mainly to make it so I don't have to manage balancing cores and can more easily choose just one file, if I choose that. Would that be any problem as far as results processing or anything goes?

Where was 2.1.5 released, or was that a mistake? 2.1.4 is the most recent I'm aware of.

With the way the stats are pulled into the NPLB database, will the bug I described here take effect (i.e. give full credit for abandoned tests)?

2.1.5 only updated the server and addresses the abandoned tests issue.

[mdettweiler]

Originally Posted by Mini-Geek

Cool, cool. I'm considering using a private PRPnet server in the future, run on my machine, mainly to make it so I don't have to manage balancing cores and can more easily choose just one file, if I choose that. Would that be any problem as far as results processing or anything goes?

That shouldn't be a problem, though you'll need to convert the results to LLR format and sort them by n primary and k secondary before sending them to Gary (since in the PRPnet results file they're returned in the order they're receved by the server). This is what we do for all of the public LLRnet and PRPnet servers so that everything is in proper order and in the same format for Gary's big stash of residuals.

Where was 2.1.5 released, or was that a mistake? 2.1.4 is the most recent I'm aware of.

With the way the stats are pulled into the NPLB database, will the bug I described here take effect (i.e. give full credit for abandoned tests)?

As Mark said--2.1.5 only contains changes to the server, and was only distributed by email (the full package is available on Mark's server here in case anyone's interested). In fact, 2.1.4 only contains changes to the server as well; all of 2.1.3 through 2.1.5 are identical client-wise. However, the most recent version of the client binaries we have posted here is 2.1.2, which does not contain the new features present in 2.1.3+. Since 2.1.5 is the latest, I'll simply use that in the client packages.

Regarding the abandoned tests bug, that shouldn't be an issue; affected versions of the server shouldn't output any actual results to the completed_tests.log file (since there's no data to use for the residual anyway). Thus, even if it messes up the PRPnet built-in stats, no bad data will go into the database. The only stuff that goes into the DB is just what's outputted to the completed_tests.log file (which is copied off daily and converted to LLRnet format for the DB to import).

Max

[Mini-Geek]

Originally Posted by rogue

2.1.5 only updated the server and addresses the abandoned tests issue.

Oh ok, but the favicon thing still isn't working. Sorry if I wasn't clear enough before, but having "http://server/favicon.ico" return '<link rel="icon" type="image/ico" href="AnyImageFileValidOrNot.ico">' will not work. Some browsers don't support non-.ico files as favicons, (Firefox, which I am using to test this, supports anything it can render) so that's partially right, but you need favicon.ico to be the .ico-format image file, not to return HTML that attempts to point the browser to a different image (which, by the way, still does not return properly - going to http://server/prpnet.ico does not work right).
i.e. when a request for favicon.ico is received, the server must return the contents of prpnet.ico, not any sort of link to prpnet.ico.

[Brucifer]

what the heck is a favicon?????

[PCZ]

A favourites icon.

[kar_bon]

Originally Posted by Brucifer

what the heck is a favicon?????

put an icon-file called 'favicon.ico' in the root-dir of the homepage and this icon will be shown in your browser favorites (here IE7) as the picture below.

[rogue]

Originally Posted by Mini-Geek

Oh ok, but the favicon thing still isn't working. Sorry if I wasn't clear enough before, but having "http://server/favicon.ico" return '<link rel="icon" type="image/ico" href="AnyImageFileValidOrNot.ico">' will not work. Some browsers don't support non-.ico files as favicons, (Firefox, which I am using to test this, supports anything it can render) so that's partially right, but you need favicon.ico to be the .ico-format image file, not to return HTML that attempts to point the browser to a different image (which, by the way, still does not return properly - going to http://server/prpnet.ico does not work right).
i.e. when a request for favicon.ico is received, the server must return the contents of prpnet.ico, not any sort of link to prpnet.ico.

If you can provide code to get this to work (see HTMLOutput.cpp), then I will incorporate it.

[Brucifer]

@Max -- The GB stats show I found a prime on G2000, however I didn't receive an email. Am I supposed to when doing the double check stuff on it now?

@PCZ re the favicon reply, thanks! :-)

[mdettweiler]

Originally Posted by Brucifer

@Max -- The GB stats show I found a prime on G2000, however I didn't receive an email. Am I supposed to when doing the double check stuff on it now?

Oh! Whoops, I forgot to turn on email notification for that port. However, come to think of it, it's probably better that it's to be left off, since right now G2000 is doing work that is both doublecheck and non-top-5000, and thus even in the somewhat remote chance that a user found a *new* prime, they still wouldn't have to do anything with it. Thus, we'll simply verify all the primes when we process the results on the backend and report any primes that were missed publicly.