Contacting users with problem machines

markhl · 2003-10-02, 03:25

People have identified error-prone machines, machines that are running v18 and keep asking for new assignments, machines that are running far slower than expected, ...
In some cases, such machines have been running for a while.

Is there some mechanism to contact users?
Friendly e-mail from George?
Pointer from Xyzzy to a relevant forum posting?

Unless they drop into a forum and happen to see their name mentioned (curtisc), they might not find out!

I appreciate that some machines and some users will be unreachable, and users may be unwilling to change. But several users would be glad to know about the issue and have the chance to help more effectively. After all they signed up to help!

GP2 · 2003-10-02, 07:12

[I moved this thread to the main Data forum]

Well you could tell curtisc that his machine "phi" is error-prone: 2 bad results, 3 good results, and 18 unverified results needing a double-check. In general though his farm is in pretty good shape (a bunch of machines I'd like to collect more stats on, though). I'm not sure if knowing it will do him any good though because it doesn't look like this machine has any current assignments.

Team_Prime_Rib has a bunch of bad machines but they already keep track of them themselves.

Unfortunately there's no way to tell the Prime95 version or CPU type for returned exponents, at least not from the status.txt and cleared.txt files. Maybe the Primenet server knows internally. If you could somehow identify all the P4s running versions pre-23 it would be very interesting, but I suspect in many large organizations GIMPS is flying under the radar and the local member doesn't have easy ongoing access to all the machines running Prime95.

By the way, the example above illustrates that there will be a lot of error-prone exponents to test: here, a single error-prone machine has generated 18 exponents that ought to be double-checked early. If you take even a modest number of error-prone machines and multiply by some reasonable average number of exponents done per machine, the numbers add up quickly.

I just e-mailed George to ask if it would be reasonable to ask him to release a couple thousand or so exponents for first-time LL testing...

2003-10-02, 03:25	#1
markhl Apr 2003 California 2²·23 Posts	Contacting users with problem machines People have identified error-prone machines, machines that are running v18 and keep asking for new assignments, machines that are running far slower than expected, ... In some cases, such machines have been running for a while. Is there some mechanism to contact users? Friendly e-mail from George? Pointer from Xyzzy to a relevant forum posting? Unless they drop into a forum and happen to see their name mentioned (curtisc), they might not find out! I appreciate that some machines and some users will be unreachable, and users may be unwilling to change. But several users would be glad to know about the issue and have the chance to help more effectively. After all they signed up to help!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Feature request - contacting server again in 67 minutes	patrik	Software	3	2006-09-25 06:37
strange problem with torture test on 16core machines	TheJudger	Hardware	5	2006-04-08 11:20
Computers not contacting Primenet	drew	Software	10	2006-02-12 18:56
For Old Users	Citrix	Prime Sierpinski Project	15	2004-08-22 16:43
Client stalls when contacting PrimeNet...	Xyzzy	Software	3	2002-08-28 21:48

2003-10-02, 07:12	#2
GP2 Sep 2003 3²·7·41 Posts	[I moved this thread to the main Data forum] Well you could tell curtisc that his machine "phi" is error-prone: 2 bad results, 3 good results, and 18 unverified results needing a double-check. In general though his farm is in pretty good shape (a bunch of machines I'd like to collect more stats on, though). I'm not sure if knowing it will do him any good though because it doesn't look like this machine has any current assignments. Team_Prime_Rib has a bunch of bad machines but they already keep track of them themselves. Unfortunately there's no way to tell the Prime95 version or CPU type for returned exponents, at least not from the status.txt and cleared.txt files. Maybe the Primenet server knows internally. If you could somehow identify all the P4s running versions pre-23 it would be very interesting, but I suspect in many large organizations GIMPS is flying under the radar and the local member doesn't have easy ongoing access to all the machines running Prime95. By the way, the example above illustrates that there will be a lot of error-prone exponents to test: here, a single error-prone machine has generated 18 exponents that ought to be double-checked early. If you take even a modest number of error-prone machines and multiply by some reasonable average number of exponents done per machine, the numbers add up quickly. I just e-mailed George to ask if it would be reasonable to ask him to release a couple thousand or so exponents for first-time LL testing...