PRPnet servers for NPLB - Page 18

mdettweiler · 2014-06-22, 06:27

My plan will be to get all the NPLB/CRUS servers upgraded to version 5.2.8 within the next couple of weeks. That version's been running stably at PrimeGrid since August, and it looks like most of the servers there are still running it. The 5.3.x releases don't add any new features relevant to us, and the changelog indicates refactoring of key code in 5.3.0, so this will give us the opportunity to observe it in action for a while before putting it into production. There's no database changes between 5.2.x and 5.3.x, so it'll be a drop-in executable replacement down the road (something I can do in <15 minutes).

Does anyone running personal servers at noprimeleftbehind.net have a desire for theirs to be upgraded to 5.2.8? Right now the personal servers are kind of all over the map - I know Gary's is at 4.1.4, and my own (which I haven't used in years) is still on 4.0.6.

I think everyone else's is at 4.3.6, but I could be wrong. If they're not and you'd like them to be upgraded, please let me know and I'll get it done promptly!

AMDave · 2014-06-22, 08:37

The change log for 5.3.1 is short on detail.

Either one of 5.3.0 or 5.2.8 should be regarded as unstable and skipped.

If the bugs that were fixed in 5.3.1 existed only in 5.3.0, then I would be ignoring any instance of 5.3.0 and watching all instances of 5.3.1
but
If the bugs that were fixed in 5.3.1 existed in 5.2.8 I'd skip to 5.3.1 since this was bug fix only, regardless of the un-used feature add in 5.3.0.

my 2c.
Your call.

mdettweiler · 2014-06-22, 18:44

Indeed. However, I would note that the WWWW-related bugs have no significance for us, since those code paths are only taken for other types of projects being done at PrimeGrid (not for "standard" LLR/PFGW projects such as ours). Also, client bugs are not an issue, since individuals will choose which client versions they run against our server anyway. That leaves:

Code:

    prpserver:  Fix an issue with length calculation of candidates.

At present, we're not doing anything with candidate length (i.e. # of digits) - our servers sort only by k and n. So again, my thinking is that this isn't going to affect us.

Mark and/or Lennart, do you know if this bug was present previously in 5.2.8, or introduced in 5.3.0?

Lennart also PMed me noting that we'll need 5.3.1 if we want to use PRPnet's doublechecking feature - so far we haven't actually made use of that, since it doesn't integrate well with our stats system (that is, it runs two tests for each candidate and reports them in the same results file - as currently structured, our DB would throw out the second as a duplicate). Our doublechecking efforts so far have been conducted as separate "first pass style" runs, the results of which will (eventually) go into a separate database for cross-checking.

My hunch is that 5.2.8 is a safe bet because it's been running stably at PrimeGrid since last August, and - in particular - they're still using it on the vast majority of their servers (the ones that run efforts similar to ours). I would assume that, if the bugs fixed in 5.3.1 affected 5.2.8, those servers would have been promptly upgraded. As noted above, most of the bugfixes in 5.3.1 are in the client; my reason for being hesitant to go to 5.3.1 on the server right away is that there was quite a bit of refactoring in the 5.3.0 release, which has the potential to introduce bugs that haven't yet been uncovered. 5.2.8, on the other hand, has a 10-month, well-proven track record.

Nonetheless, I'll wait to get a definitive answer from Mark and/or Lennart (since admittedly, I haven't been following the recent releases closely).

mdettweiler · 2014-06-22, 19:29

I found the length-calculation bugfix in Sourceforge (revision 17):

Click image for larger version

Name: r17.PNG
Views: 130
Size: 21.4 KB
ID: 11360

This is extremely minor - just a missing floor() on the decimal length calculations. Again, since we're not using length calculations for anything important, my thinking is that this isn't a big deal.

I'm continuing to look back through the Sourceforge changelogs (the code was first put into Sourceforge at a prerelease v5.3.0 state) - but, I think this is the only server-side change relevant to us.

Edit: I've finished going through all the changelogs since the initial import to Sourceforge. The only other change I found that was relevant to us is a minor fix in the mail sending code; I wonder if that would get mail working again on our servers? (I know I haven't been getting any PRPnet mails for a long time, but last I recall other people were getting them fine, so it may just be me.)

rogue · 2014-06-22, 21:45

Quote:

Originally Posted by mdettweiler

I found the length-calculation bugfix in Sourceforge (revision 17):
Attachment 11360
This is extremely minor - just a missing floor() on the decimal length calculations. Again, since we're not using length calculations for anything important, my thinking is that this isn't a big deal.

I'm continuing to look back through the Sourceforge changelogs (the code was first put into Sourceforge at a prerelease v5.3.0 state) - but, I think this is the only server-side change relevant to us.

Edit: I've finished going through all the changelogs since the initial import to Sourceforge. The only other change I found that was relevant to us is a minor fix in the mail sending code; I wonder if that would get mail working again on our servers? (I know I haven't been getting any PRPnet mails for a long time, but last I recall other people were getting them fine, so it may just be me.)

There is another issue in the code for sending e-mail that I haven't been able to track down. It only happens with some SMTP servers.

That being said, I strongly recommend 5.3.1 over older versions because I won't patch 5.2.8 if it has any other bugs. You also know my track record in that I fix things quickly. And since there are zero database changes between 5.2.8 and 5.3.1 you can quickly replace the prpserver executable with a different version if a bug does crop up that you can't live with.

odicin · 2014-06-23, 18:30

Quote:

Originally Posted by gd_barnes

Max informed that he has upgraded port 9000 to 4.3.6. I'm OK with upgrading all NPLB/CRUS PRPnet servers to 5.x.x at this point but that will have to wait until Max has time.

Max, please keep port 12000 in mind. It's also an old 4.1.4 Version and keep in mind, it was on your todo list ;)

Quote:

Originally Posted by mdettweiler

Port 12000 is coming soon--I'll try to get the server upgrade in as well if I can.

Update to 4.3.6 will help a lot, because of the much better navigation.

Regards Odi

mdettweiler · 2014-06-26, 04:04

Quote:

Originally Posted by odicin

Max, please keep port 12000 in mind. It's also an old 4.1.4 Version and keep in mind, it was on your todo list ;)

Update to 4.3.6 will help a lot, because of the much better navigation.

Regards Odi

Ah, yes - thank you for the reminder! I do tend to forget about those ports since they went for a while without having work loading in them, but, indeed, they are very much active now. I'll be sure to upgrade it as well - first to 4.3.6, then the rest of the way to 5.2.8 with the others.

Lennart · 2014-06-28, 22:20

Just noticed that I can't reach any prpnetservers at NPLB or CRUS.

Lennart

mdettweiler · 2014-06-29, 01:33

Yeah, me neither. I can't reach the http://noprimeleftbehind.net/ website either - looks like a general server outage.

I'll cross-post in the "Server outrages" thread.

gd_barnes · 2014-09-11, 09:22

Max,

Please change the subtitle for port 1468 on http://www.noprimeleftbehind.net/prpnet to show (k=300-400 17k maxi-drive).

Thanks,
Gary

mdettweiler · 2014-09-11, 11:30

Quote:

Originally Posted by gd_barnes

Max,

Please change the subtitle for port 1468 on http://www.noprimeleftbehind.net/prpnet to show (k=300-400 17k maxi-drive).

Thanks,
Gary

Done.

2014-06-22, 08:37	#189
AMDave Jan 2006 deep in a while-loop 292₁₆ Posts	The change log for 5.3.1 is short on detail. Either one of 5.3.0 or 5.2.8 should be regarded as unstable and skipped. If the bugs that were fixed in 5.3.1 existed only in 5.3.0, then I would be ignoring any instance of 5.3.0 and watching all instances of 5.3.1 but If the bugs that were fixed in 5.3.1 existed in 5.2.8 I'd skip to 5.3.1 since this was bug fix only, regardless of the un-used feature add in 5.3.0. my 2c. Your call. Last fiddled with by AMDave on 2014-06-22 at 08:42 Reason: edited for clarity

2014-06-22, 18:44	#190
mdettweiler A Sunny Moo Aug 2007 USA (GMT-5) 14151₈ Posts	Indeed. However, I would note that the WWWW-related bugs have no significance for us, since those code paths are only taken for other types of projects being done at PrimeGrid (not for "standard" LLR/PFGW projects such as ours). Also, client bugs are not an issue, since individuals will choose which client versions they run against our server anyway. That leaves: Code: prpserver: Fix an issue with length calculation of candidates. At present, we're not doing anything with candidate length (i.e. # of digits) - our servers sort only by k and n. So again, my thinking is that this isn't going to affect us. Mark and/or Lennart, do you know if this bug was present previously in 5.2.8, or introduced in 5.3.0? Lennart also PMed me noting that we'll need 5.3.1 if we want to use PRPnet's doublechecking feature - so far we haven't actually made use of that, since it doesn't integrate well with our stats system (that is, it runs two tests for each candidate and reports them in the same results file - as currently structured, our DB would throw out the second as a duplicate). Our doublechecking efforts so far have been conducted as separate "first pass style" runs, the results of which will (eventually) go into a separate database for cross-checking. My hunch is that 5.2.8 is a safe bet because it's been running stably at PrimeGrid since last August, and - in particular - they're still using it on the vast majority of their servers (the ones that run efforts similar to ours). I would assume that, if the bugs fixed in 5.3.1 affected 5.2.8, those servers would have been promptly upgraded. As noted above, most of the bugfixes in 5.3.1 are in the client; my reason for being hesitant to go to 5.3.1 on the server right away is that there was quite a bit of refactoring in the 5.3.0 release, which has the potential to introduce bugs that haven't yet been uncovered. 5.2.8, on the other hand, has a 10-month, well-proven track record. Nonetheless, I'll wait to get a definitive answer from Mark and/or Lennart (since admittedly, I haven't been following the recent releases closely).

2014-06-22, 19:29	#191
mdettweiler A Sunny Moo Aug 2007 USA (GMT-5) 14151₈ Posts	I found the length-calculation bugfix in Sourceforge (revision 17): This is extremely minor - just a missing floor() on the decimal length calculations. Again, since we're not using length calculations for anything important, my thinking is that this isn't a big deal. I'm continuing to look back through the Sourceforge changelogs (the code was first put into Sourceforge at a prerelease v5.3.0 state) - but, I think this is the only server-side change relevant to us. Edit: I've finished going through all the changelogs since the initial import to Sourceforge. The only other change I found that was relevant to us is a minor fix in the mail sending code; I wonder if that would get mail working again on our servers? (I know I haven't been getting any PRPnet mails for a long time, but last I recall other people were getting them fine, so it may just be me.) Last fiddled with by mdettweiler on 2014-06-22 at 19:35

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
PRPnet Servers for CRUS	MyDogBuster	Conjectures 'R Us	83	2021-01-28 10:08
LLRnet servers for NPLB	kar_bon	No Prime Left Behind	1343	2014-08-20 09:38
Public PRPNet Servers	rogue	Open Projects	26	2013-01-16 01:33
PRPNet servers down?	opyrt	Prime Sierpinski Project	13	2009-11-04 21:33
Servers for NPLB	gd_barnes	No Prime Left Behind	0	2009-08-10 19:21

2014-06-22, 06:27	#188
mdettweiler A Sunny Moo Aug 2007 USA (GMT-5) 3·2,083 Posts	My plan will be to get all the NPLB/CRUS servers upgraded to version 5.2.8 within the next couple of weeks. That version's been running stably at PrimeGrid since August, and it looks like most of the servers there are still running it. The 5.3.x releases don't add any new features relevant to us, and the changelog indicates refactoring of key code in 5.3.0, so this will give us the opportunity to observe it in action for a while before putting it into production. There's no database changes between 5.2.x and 5.3.x, so it'll be a drop-in executable replacement down the road (something I can do in <15 minutes). Does anyone running personal servers at noprimeleftbehind.net have a desire for theirs to be upgraded to 5.2.8? Right now the personal servers are kind of all over the map - I know Gary's is at 4.1.4, and my own (which I haven't used in years) is still on 4.0.6. I think everyone else's is at 4.3.6, but I could be wrong. If they're not and you'd like them to be upgraded, please let me know and I'll get it done promptly!

2014-06-28, 22:20	#195
Lennart "Lennart" Jun 2007 2140₈ Posts	Just noticed that I can't reach any prpnetservers at NPLB or CRUS. Lennart

2014-06-29, 01:33	#196
mdettweiler A Sunny Moo Aug 2007 USA (GMT-5) 3·2,083 Posts	Yeah, me neither. I can't reach the http://noprimeleftbehind.net/ website either - looks like a general server outage. I'll cross-post in the "Server outrages" thread.

2014-09-11, 09:22	#197
gd_barnes May 2007 Kansas; USA 2×5,153 Posts	Max, Please change the subtitle for port 1468 on http://www.noprimeleftbehind.net/prpnet to show (k=300-400 17k maxi-drive). Thanks, Gary