Help me test the new client

[jjjjL]

Can you guys download and test out these new "packaged" versions of the client?

They are just the base client with a prime.txt file already embedded so it will use SB's server. All you need to do now is open the prime.txt file and change username from "louie" to your SB username.

WINDOWS - http://www.seventeenorbust.com/download/p95v259.zip
WINDOWS x64 - http://www.seventeenorbust.com/download/p64v259.zip
LINUX - http://www.seventeenorbust.com/download/mprime259.zip
LINUX x64 - http://www.seventeenorbust.com/downl...59-linux64.zip
Mac OSX - http://www.seventeenorbust.com/downl...259-MacOSX.zip
Mac OSX x64 - http://www.seventeenorbust.com/downl...9-MacOSX64.zip

These are better than the original instructions because it won't download GIMPS tests first and pull the work out of their system. It also will work fine if someone doesn't update the prime.txt file at all. And I changed the server so later you can change the username and make it your own if you forget to update it from my username.

Are there any changes I should make before I make these files available to download for the average user? Does the windows client work well? Does it auto-start on reboot? Linux? Mac OS X? Anything you can think of would help. Thanks.

Cheers,
Louie

[enderak]

Getting a 404 Not Found on the windows versions... Looks like the "w" in the file name should be a "v".

EDIT: Got the packaged version downloaded, and it works fine - the only problem I see is that prime.txt uses the Unix line separator instead of DOS, so there are no line breaks when opened in Notepad. Opens fine in Wordpad or more advanced text editors, but could be confusing for folks accustomed to using Notepad.

I think the client works well, including auto-start (Windows version at least - didn't test the others) so if you think everything is solid server-side, I say go for it

[DukeBox]

Are there any changes I should make before I make these files available to download for the average user? Does the windows client work well? Does it auto-start on reboot? Linux? Mac OS X? Anything you can think of would help. Thanks.

Why not add a batch/sh file that alters the prime.txt ? That's pretty easy to create, I made it also for multisob.

[Matt]

32-Bit linux version seems to be working, all i did was change the username and run it. It's connected to the server and got a test.

[jjjjL]

Why not add a batch/sh file that alters the prime.txt ? That's pretty easy to create, I made it also for multisob.

I thought about going this route but I think this may be easier.

http://www.seventeenorbust.com/account/prime.txt.mhtml

Anyone logged into the site can get their prime.txt file there. Work for everyone?

Does anyone know a way for me to make a button that would download the text box as a file?

Either way, I think this is a good middle-ground for those who don't want to follow instructions to edit the file themselves.

Cheers,
Louie

[DukeBox]

Does anyone know a way for me to make a button that would download the text box as a file?

That depends on the mime header of the link. Since you have php, you can create a link that sends other headers for the file so it's offered as an raw txt file.

Anyhow, i cannot send and get any work, http://www.seventeenorbust.com/v5server/index.php gives back that there are to many connections to the mysql database. Probably the same that's going on on the main site.
[fixed]

[Death]

it should be like onclick:javascript.location=file.txt

something like that.

[Matt]

<?php
header('Content-type: text/plain');
header('Content-Disposition: attachment; filename="prime.txt"');
?>

Those are the headers you would need, then just the code to generate the prime.txt which includes the person's username.

[jjjjL]

i cannot send and get any work, http://www.seventeenorbust.com/v5server/index.php gives back that there are to many connections to the mysql database. Probably the same that's going on on the main site.

System error

error:  DBI connect('sb:localhost','sbweb',...) failed: Too many connections at /usr/local/apache/htdocs/sb/db/connect.mc line 11 

Maybe that was earlier today when I was added the k=21181 tests to the queues. There was a weird bug in the factor removal code and I had to add all the tests to the queues and then delete all the factored ones manually unlike the other k.

I don't see any problems like that now. Let me know if you see it again. It shouldn't have been down for more than 15 minutes.

Cheers,
Louie

[jjjjL]

<?php
header('Content-type: text/plain');
header('Content-Disposition: attachment; filename="prime.txt"');
?>

Those are the headers you would need, then just the code to generate the prime.txt which includes the person's username.

Yeah, that code will work. Thanks!

There's a link at the bottom of
http://www.seventeenorbust.com/account/prime.txt.mhtml now

I'll probably add a link like that for those who are logged in and have instructions to either login and download that or edit the prime.txt file by hand when I roll-out the download for regular users.

Cheers,
Louie

[jjjjL]

Good news! George Woltman is removing a minor bug I found in Prime95 that prevents optimal P-1 factoring of PRP tests.

Now when we get to ~17.2M we can start doing P-1 testing in the client.


Cheers,
Louie

[jjjjL]

Download page for the new client is ready to go but not live yet....

http://www.seventeenorbust.com/download/2index.mhtml

Everything look good?

I'm considering launching it soon but I just added some integrity check tests into the pool to make sure the Prime95 server / clients are really behaving well and giving expected results. Once these sanity checks complete, I'll likely launch this new option for people unless we run into difficulties.

One thing that concerns me is that when I've used the windows client a couple times, I was able to accidentally blow away the "MersenneIP=www.seventeenorbust.com" line by opening the Preferences menu. Anyone know exactly what causes this and what counter-measures could be taken to avoid this loss of a client? Is there a better way to specify the line in the prime.txt file or a way to set a certain setting so it won't do that? Let me know if anyone figures this out.

And give me feedback if you see an improvement that could be made to the new download page before it becomes live.

Cheers,
Louie

[IronBits]

Too much in the .zip files.
I took the 64bit client, found MAC stuff in there.
There is also a prime.txt file in it.
the prime.txt file for windows doesn't contain crlf so looks like one continuous line in notepad.

[umccullough]

Louie,

I took a peek at the code from the gimps FTP in an attempt to port it to Haiku...

I made all the necessary changes to get the code compiling on Haiku, but when linking, the linux gwnum.a elf was unlinkable for me...

Note, this was actually a couple weeks ago, and I basically have formatted that machine since then - but the patches were pretty minor, I can re-create them pretty easily I suspect.

Am I to understand that the platforms available for this are basically limited to whatever objconv supports at this point? Is there a version of gwnum sources floating around that were written in something other than masm?

If you're willing to chat via email, let me know - I'd love to get a client for Haiku on the map

- Urias
Team Haiku

[DukeBox]

I'm also very curious if it's possible to compile the client for GPU's.

[enderak]

I don't know if you still want us to post completed tests, or if you are just monitoring all results n>16M, but my first 4 first-pass tests on the new client completed last night.. Results are below.

[Fri Mar 06 21:07:23 2009]
UID: enderak, 22699*2^16000534+1 is not prime.  RES64: 761D2BC0BC851BFD. Wd4: 985E7085,00000000, AID: 000000000000000000000000000ED3E1
[Fri Mar 06 21:28:39 2009]
UID: enderak, 24737*2^16000591+1 is not prime.  RES64: D9A6F876B46147EB. Wd4: 98D070F7,00000000, AID: 000000000000000000000000000ED3E2
[Fri Mar 06 22:32:56 2009]
UID: enderak, 10223*2^16000457+1 is not prime.  RES64: 8BB11BA80D7FEB3A. Wd4: 97C46FEB,00000000, AID: 000000000000000000000000000ED3E0
[Fri Mar 06 23:28:43 2009]
UID: enderak, 24737*2^16000423+1 is not prime.  RES64: CE4419B347CBC4E9. Wd4: 97806FA7,00000000, AID: 000000000000000000000000000ED3DF 

[enderak]

I'm not 100% sure about this but it seems like if you reserve a test early due to queuing up work, when it actually starts crunching the test, the CEMs/s value for those tests is computed from the time you reserved them, so you end up starting with very low cEMs/s and slowly increases over time. I should be getting 8.5M+ cEMs/s for these tests, but this morning they were around 1.5M and now they are at 4.5M

If this is in fact what's happening, is it possible to calculate the cEMs/s starting from the time the first communication with the server where work was actually done on the test?

EDIT: I was thinking you could just set DaysOfWork=0 in the prime.txt that is generated on the download page, but this doesn't seem to work. It just says there is no work to do.

[jjjjL]

Louie,

I took a peek at the code from the gimps FTP in an attempt to port it to Haiku...

I made all the necessary changes to get the code compiling on Haiku, but when linking, the linux gwnum.a elf was unlinkable for me...

Note, this was actually a couple weeks ago, and I basically have formatted that machine since then - but the patches were pretty minor, I can re-create them pretty easily I suspect.

Am I to understand that the platforms available for this are basically limited to whatever objconv supports at this point? Is there a version of gwnum sources floating around that were written in something other than masm?

If you're willing to chat via email, let me know - I'd love to get a client for Haiku on the map

- Urias
Team Haiku

1) Doesn't the source include a pre-compiled gwnub.lib instead of the .a file? Does that work any better?

2) Remember, you don't have to have a working objconv FOR Haiku, you only need one on a computer that can TARGET Haiku.... so if you just know the file-type that you need it changed to, a linux box should be enough to convert the file for you.


Cheers,
Louie

[jjjjL]

Too much in the .zip files.
I took the 64bit client, found MAC stuff in there.
There is also a prime.txt file in it.
the prime.txt file for windows doesn't contain crlf so looks like one continuous line in notepad.

Good call. I hate it when I decompress things and it has those extra files.

I ran a program that clears those out on the zip files and also converted the prime.txt to DOS format linebreaks.

Let me know that both changes worked, please.

I wonder if I should re-compress the linux versions using tar.gz so the permissions work correctly? Let me know if this will help.


Cheers,
Louie

[Fozzie]

How do you compare the performance of the old client with the new one.

It seems to be taking a hell of a long time to complete a set of tests.

[jjjjL]

I'm not 100% sure about this but it seems like if you reserve a test early due to queuing up work, when it actually starts crunching the test, the CEMs/s value for those tests is computed from the time you reserved them, so you end up starting with very low cEMs/s and slowly increases over time. I should be getting 8.5M+ cEMs/s for these tests, but this morning they were around 1.5M and now they are at 4.5M

If this is in fact what's happening, is it possible to calculate the cEMs/s starting from the time the first communication with the server where work was actually done on the test?

EDIT: I was thinking you could just set DaysOfWork=0 in the prime.txt that is generated on the download page, but this doesn't seem to work. It just says there is no work to do.

You're right. cEMs for reserved tests have been using their reservation time as their start time, not their actual start time in the client.

I've changed the code now so that if a test's progress == 0, it will reset the starting time. I think this will fix this behavior because my assumption is that the client reports the completion time with a 0% progress report at the time of first reservation AND again right before it begins work on each new test. If my assumption is correct, my update to the server should fix this.


Any tests you've reported progress on so far will still have the timing behavior you noticed. But let me know if future tests behave this way too or if the problem is fixed by my change. Thanks!

Cheers,
Louie

[jjjjL]

It would be nice if this table could be updated: http://www.seventeenorbust.com/stats/byOS.mhtml

Adding in XP 64bit as well as Vista 32 and 64bit.
And while you are there, you might as well plug in Windows 7

Ok, done.


Cheers,
Louie

[jjjjL]

I'm also very curious if it's possible to compile the client for GPU's.

There are folks exploring this on other forums:
http://episteme.arstechnica.com/eve/...m/744004804931

I downloaded the CUDA SDK and tried compiling GIMPS.
http://www.nvidia.com/object/cuda_get.html

By try, I mean complied the examples, installed libraries, opened the Prime95 source code, opened the CUDA templates, cried, and gave up. But compiling glucas may not be as hard. Other efforts to do that like the one above appear to have stalled with no progress though.

Maybe you could try to contact these people who seem to be doing something similar:
http://forums.nvidia.com/lofiversion...hp?t70483.html


I'm open to supporting someone in porting a client but only if they are a developer and they plan to invest some real time and effort into this.

Another option would be to produce a sieve client. That might be much more fruitful in the near term. I could imagine a sieve client being incredibly powerful and much easier to parallelize.

Cheers,
Louie

[umccullough]

1) Doesn't the source include a pre-compiled gwnub.lib instead of the .a file? Does that work any better?

I haven't had a chance to experiment with this much further - I'll admit that the intricacies of linking are a little foreign to me

I'll post my failure output somewhere in the next couple days.

2) Remember, you don't have to have a working objconv FOR Haiku, you only need one on a computer that can TARGET Haiku.... so if you just know the file-type that you need it changed to, a linux box should be enough to convert the file for you.

I'll look into this - I know Haiku uses elf, but I think it's got some custom header info that maybe isn't compatible with linux. I had assumed this was part of my problem, but perhaps it isn't

[jjjjL]

I just upgraded SB's Primenet server to allow for in-client un-reserving.

It's a small tweak in some ways but every little bit helps. This means if someone uses the "Quit GIMPS" option it will at least un-reserve their tests right away instead of 30 days after the last report.

And I think the client may unreserve tests on it's own from time to time in other situations like when it falls behind.

The website's un-reserve page in the "Preferences" section will still work too if you need it during testing.

Hope this helps.

Cheers,
Louie

[riptide]

How do you compare the performance of the old client with the new one.

Bump! +1. I would like to know this aswell!

[enderak]

If you look at the sb.log file, you can figure out the difference in time (in seconds) over the difference in n.high and get the sec/iteration value that you can directly compare with what the GIMPs client outputs.

In general, I think the new client typically runs around 10-15% faster.

[jjjjL]

OK, almost ready to release these clients to the whole user pool. But first, I need some major feedback. Try to break things people.

Can everyone testing please download the new client for your system from http://www.seventeenorbust.com/download/2index.mhtml

I know the filenames are similar but the client's binaries have 2 important changes I want everyone to have. It enables P-1 factoring (although it won't be actively used until we reach n > 17.2M) and has safer save-file naming. Both changes come courtesy of George Woltman who has been working overtime to get us these specific changes so we can start safely and efficiently using his client. Very cool stuff.

So if you have current work, you don't need to update any txt files from this update. Just replace the main executable.

However, if you haven't tested yet, or are using a new computer, just follow the instructions on the above download page. What do you think? Clear enough to setup?

Let me know if you run into any problems.

Cheers,
Louie

[enderak]

It appears that on my 64-bit machine it is doing a P-1 test, but my 32-bit machines skipped it.

How does the P-1 factoring determine how much RAM to use? On my i7 w/ 12GB memory (Vista 64-bit), it says P-1 would use "up to 4098MB of memory", but on my older Xeon w/ 4GB memory (XP 32-bit), it says it would only use 8MB.

Only using 8MB seems like a miniscule amount of memory when I would think it could just as easily and effectively use quite a bit more.

Other than the memory question, I updated my two machines and installed it fresh on a Windows Server 2003 machine and all seems to be working OK.

[jjjjL]

It appears that on my 64-bit machine it is doing a P-1 test, but my 32-bit machines skipped it.

How does the P-1 factoring determine how much RAM to use? On my i7 w/ 12GB memory (Vista 64-bit), it says P-1 would use "up to 4098MB of memory", but on my older Xeon w/ 4GB memory (XP 32-bit), it says it would only use 8MB.

Only using 8MB seems like a miniscule amount of memory when I would think it could just as easily and effectively use quite a bit more.

Other than the memory question, I updated my two machines and installed it fresh on a Windows Server 2003 machine and all seems to be working OK.

That's interesting that it decided to do a P-1 test at all. I had the param set to saved=0.5 but with the factor limit up to 2^55, it seems like it should never do a test. Anyway, I've moved it back down to 0.0 so there really shouldn't be any P-1 testing until n > 17.2 now.

It could use more memory potentially. It makes the P-1 Step 2 more efficient. It's nice to have, but not a huge deal. I'd rather the client was conservative on how much memory it bloat into on casual users systems. People who really want to find factors (and get the 2x the test cem credit), will likely raise that limit just like you have.

Thanks for the early feedback.

Cheers,
Louie

[enderak]

Ah, I forgot I changed the memory settings in Prime95 on this machine - in fact I forgot that setting was even there. I went ahead and updated my old machine to allow 1GB to be used.

It looks like for some reason the factor limit was at 2^54 and saved=1.5... (log below) I'll keep an eye out when this machine starts its next tests and see what it does.

[Mar 16 11:49] Optimal P-1 factoring of 21181*2^16018652+1 using up to 4096MB of memory.
[Mar 16 11:49] Assuming no factors below 2^54 and 1.5 primality tests saved if a factor is found.
[Mar 16 11:49] Optimal bounds are B1=35000, B2=271250
[Mar 16 11:49] Chance of finding a factor is an estimated 0.335%
[Mar 16 11:49] Using all-complex FFT length 1536K, 2 threads

[tqft]

Setup: Ubuntu 8.10 x86

I have this scipt setup to start sb
gnome-terminal --working-directory=/home/ian/sb3 --command="./sb sclient.conf" --title=sb3

I have changed this to
gnome-terminal --working-directory=/home/ian/sb3 --command="./mprime sclient.conf" --title=sb3-mprime

The only executable in the linux download is named mprime - correct? You will need to update the user instructions to reflect this or change the executable name. Or something else of you choice.

Should I be seeing any output at all? Because I am not yet (but haven't waited long)

[jjjjL]

Setup: Ubuntu 8.10 x86

I have this scipt setup to start sb
gnome-terminal --working-directory=/home/ian/sb3 --command="./sb sclient.conf" --title=sb3

I have changed this to
gnome-terminal --working-directory=/home/ian/sb3 --command="./mprime sclient.conf" --title=sb3-mprime

The only executable in the linux download is named mprime - correct? You will need to update the user instructions to reflect this or change the executable name. Or something else of you choice.

Should I be seeing any output at all? Because I am not yet (but haven't waited long)

Ah, I see the confusion.

This software is not a "drop-in" replacement for previous SB clients. This is brand new software. So even though you're a current user, you still have to follow the instructions in "Step 3" on http://www.seventeenorbust.com/download/2index.mhtml

And you should not use that same command line. mprime takes no arguments. The Mac/Linux clients don't output data in the console. If you want to see what it's doing, tail -f results.txt will give you occasional updates. The Windows client is more verbose in the client window itself.


Cheers,
Louie

[tqft]

So how are you recommending people change over? Let old clients complete work in progress, expire new test then start new client.

[jjjjL]

So how are you recommending people change over? Let old clients complete work in progress, expire new test then start new client.

Yes, that sounds terrific. I may just recommend people upgrade in general but I could remind people who are committed users to go the extra mile and complete their last test if possible.


Cheers,
Louie

[jjjjL]

I'm gonna close this thread since now this client is released and this thread was meant for testing before release.

Post here if you want to help trouble-shoot now that the new client is released:
http://www.free-dc.org/forum/showthread.php?t=18264

Cheers,
Louie