v5 PrimeNet alpha & beta testing - Page 3

Yxine · 2006-02-10, 03:49

I have 5+ computers available and like to participate in beta-testing!

moo · 2006-02-10, 06:38

Quote:

Originally Posted by Old man PrimeNet

I have a copy of Prime95 14.4 for moo, just in case v5 blows up his computer.

whats this all about.... im saying it gives me a fall back aka might try to purpose it to my school which has about 80 or 90 prescots some early ht machines and a few hundred p3s this gives me a fallback incase they decide they dont want it -_- then i can say with a press of da button i can remove it from every machine.

Paulie · 2006-02-13, 00:48

Quote:

Originally Posted by Old man PrimeNet

Teams will be managed entirely on the server's web site. A computer is the v5 identity 'atom', though individual CPUs can be assigned different kinds of work. A computer's owner is established when the client first registers the computer with v5, but thereafter ownership is managed entirely on the server's web site, similar to teams.

This scheme maximizes flexibility and convenience while tightening and fixing control, ownership and admin issues we have with v4 accounts and teams.

Sounds yummy.

So for those with dozens (hundreds) of machines, how will they control them? Individual login per computer, or will they all be under a single interface? How is "team" membership going to be done, under the computer account management page, you select your team? And with your team identified, you're then allowed to share work within the team?

Just trying to picture how things will work in comparison to the existing team framework.

Old man PrimeNet · 2006-02-13, 09:20

Two years ago this design document initially captured the v5 system concept and objectives, though it's now woefully outdated and lacks details for feature-level implementation. However, the v5 GIMPS client web API specification is current for the system in development and describes data-level dependencies that hint at how the system works and what will be available for users on the web site.

A single web interface (a set of forms/pages) will manage all the computers of an account, as a whole or specific computers, and its operating options and preferences. Another set will manage all the resources of a team, which is created and managed by a teamleader. An important design philosophy of v5 is allowing users to manage distributed computing resources across multiple projects and on various workloads without having to physically revisit their computers once they have installed the v5 client.

Before you can install your first v5 client program, you normally need to create an account on the web site. This sets up an account summary page you can optionally make public and to which you can link or frame-embed from elsewhere, and enables your access to manage your account, computers and workload. It also provides an account binding the v5 client expects when you install it, though it's also possible to install a v5 client with an anonymous account binding that is instead managed by the project admin (in GIMPS' case, that's big GW).

When you install a v5 client, you give it just your account name, an optional computer name, and nothing else! This can be setup for batch installs, too. Once installed the v5 client does not need to keep a record of who owns it or its name, just its identity key. From that point forward the v5 server tells the client, when it asks, who owns it and what its name is.

Installed v5 clients will appear on your personal admin web page. You can still manage the client directly at the computer to some degree (including changing its account owner and local options and work preferences), but for the vast majority of the time you will select your project and work-type preference options on the web and your group (or cloud) of computers will simply perform your wishes on their own, collectively, or even individually or even as different CPUs on a computer if you want to fine-tune them. You will also be able to rename computers and transfer them between accounts. The local v5 client settings, however, will always have the ability to over-ride your web account (or team) settings.

You choose your optional team on your admin web page and can change it as often as you wish - your computers ride along with you. We haven't finalized decisions about team-level computer control, but we are carefully thinking it through. The current plan for crediting work in teams is that whatever team-level credit you earn while you are part of a team stays with the team. The same plan applies to computers on accounts - you can give away or lose the computer but the work credit it earned for your account stays on your account.

Much of the alpha & beta test phases will be focused upon these web management tasks, teams, and stats reporting rather than the v5 client and server. For alpha in particular we don't need a ton of computers per person - only enough to work out the kinks until beta. At first the v5 web site will be a little rough around the edges until the functionality is firm, but we'll pretty it up later. Get it working, then tune it up.

delta_t · 2006-02-13, 17:08

Any changes to the reports? Just checking because I've been archiving the reports since May 04 (http://tuatara.sdsu.edu/primenet/) and would like to know if there will be any changes to the reports (either names and/or locations). If so I'd like to make sure the automated scripts on my server are up-to-date with any changes.

Also, I'm game for both alpha/beta testing too. Have FreeBSD clients if that makes any difference.

ric · 2006-02-21, 13:11

Just out of curiosity

: how will the v5 server manage the case of an account with a number of unconnected machines, with a single one acting as a front-end (both for collecting new exponents and reporting results)?

TIA
ric

(2nd post on the subject, was wandering where it had finished, when I read about last week's HD failure)

Old man PrimeNet · 2006-02-21, 16:20

PrimeNet v5 will limit assignments to a single computer to some rational threshold based upon its speed, work history and results reliability because the amount of work a computer can per unit time can be very accurately determined as a function of the exponent. This will probably make using a single computer as a front-end for others off-net impractical, though undoubtedly some will try anyway.

Others may study the PrimeNet 5.0 GIMPS APIs and write scripts to emulate a client program, however the security features will treat these as open-source untrusted clients, which are subject to tighter safeguards and risk being locked out completely in the event of counterproductive activity. They also are metered to rational rate limits.

An important objective of the new system is to enable users to manage their group of computers from the server's web site and within the wider guidelines of a project's general goals, rather than resorting to sneakernetting data from computer to computer. The server will be much more selective about what work to give computers to achieve the best results within the user's configured options, so there's less reason to invent strategies to get better assignments.

Overall, we would rather add any needed fine-tuning workflow features to the v5 server than encourage script-writing and sneakernetting to achieve the same effective result outside of the server's workflow visibility. Defeating PrimeNet's assignment system for personal reasons (climbing stats, etc.) will be less efficient for the project, if not simply counterproductive. Ideally, we want every computer participating to run a v5 client on the Internet - but we also realize that's not going to work for a few folks with special network issues or client programs without PrimeNet support. Bottom line: the easiest way to successful participation will be to use a trusted v5 client program or an open-source untrusted v5 client code.

delta_t: The reports are likely to change somewhat, but at a minimum (early on) the same information will be presented. We plan to add several new statistical reports but this will likely be among the last progressive stages of the upgrade.

PhilF · 2006-02-21, 18:21

Quote:

Originally Posted by Old man PrimeNet

An important objective of the new system is to enable users to manage their group of computers from the server's web site and within the wider guidelines of a project's general goals, rather than resorting to sneakernetting data from computer to computer. The server will be much more selective about what work to give computers to achieve the best results within the user's configured options, so there's less reason to invent strategies to get better assignments.

Here's my method, and I'd like your input to see if this strategy will still work: First, I try to get assignments that someone else has trial/P-1 factored, but has expired. But when I do get exponents that need factoring, I move them from the computer that received the assignment to one that is optimized for factoring (slower CPU, but lots of memory). If no factor is found, it is then moved again to a computer running low on assignments, which is not necessarily the computer that originally received the assignment.

I have automated scripts that help me with this process, and I believe it does help my overall throughput, which in turn helps the project's throughput.

Will this strategy still work? If not, will there be a way for me to use the server's management functions to keep my factoring machine / LL testing machines working as they are now?

BTW, thanks for all your hard work on the new server!

Old man PrimeNet · 2006-02-21, 20:18

We plan to allow computers on the same account to return results for assignments issued to each other, and for credit if the assignment key comes along.

However it is much easier for us to add a new work_preference "First-time LL w/o factoring" and then you would instead set that as your configured preference for whichever computers you want. Similarly for factoring-only computers.

Here's the current list of v5 work_preference options:

what_makes_sense - server decides
fact_trial_LMH
fact_trail_sieve
fact_pm1_small
fact_pm1_large
fact_ecm_small
fact_ecm_fermat - future
fact_ecm_cunningham - future
LL_first
LL_check
LL_world_record
LL_10M
LL_100M

I just put this new "LL_first_sans_fact" work_preference onto our unofficial "v5 Requests TO DO" list for later.

If there are other script-driven strategies like these, I'd like to learn about them so we can develop v5 requirements to eliminate script use where possible.

gribozavr · 2006-02-21, 20:39

fact_trial_LMH -- what range of exponets will be factored? Will 100mdpp fall into this?

Since there is LL_100M, why not add fact_trial_100M to avoid confusion?

tha · 2006-02-21, 20:47

Great, how about:

LL_10M_check

2006-02-13, 17:08	#27
delta_t Nov 2002 Anchorage, AK 3·7·17 Posts	Any changes to the reports? Just checking because I've been archiving the reports since May 04 (http://tuatara.sdsu.edu/primenet/) and would like to know if there will be any changes to the reports (either names and/or locations). If so I'd like to make sure the automated scripts on my server are up-to-date with any changes. Also, I'm game for both alpha/beta testing too. Have FreeBSD clients if that makes any difference. Last fiddled with by delta_t on 2006-02-13 at 17:09

2006-02-21, 13:11	#28
ric Jul 2004 Milan, Ita 2·7·13 Posts	V5 and assignments managed from a single front-end Just out of curiosity : how will the v5 server manage the case of an account with a number of unconnected machines, with a single one acting as a front-end (both for collecting new exponents and reporting results)? TIA ric (2nd post on the subject, was wandering where it had finished, when I read about last week's HD failure)

2006-02-21, 20:18	#31
Old man PrimeNet Jan 2003 Altitude>12,500 MSL 101 Posts	We plan to allow computers on the same account to return results for assignments issued to each other, and for credit if the assignment key comes along. However it is much easier for us to add a new work_preference "First-time LL w/o factoring" and then you would instead set that as your configured preference for whichever computers you want. Similarly for factoring-only computers. Here's the current list of v5 work_preference options: what_makes_sense - server decides fact_trial_LMH fact_trail_sieve fact_pm1_small fact_pm1_large fact_ecm_small fact_ecm_fermat - future fact_ecm_cunningham - future LL_first LL_check LL_world_record LL_10M LL_100M I just put this new "LL_first_sans_fact" work_preference onto our unofficial "v5 Requests TO DO" list for later. If there are other script-driven strategies like these, I'd like to learn about them so we can develop v5 requirements to eliminate script use where possible.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
PrimeNet API Manual testing or GPU machine GUID	Mini-Geek	PrimeNet	7	2015-12-23 20:54
Off to Alpha Centauri Bb ! ?	Uncwilly	Lounge	66	2012-10-24 17:36
Primenet gave me a known prime for LL testing	Mr. P-1	PrimeNet	3	2011-02-20 09:38
v5 alpha and beta testing game plan	Prime95	PrimeNet	77	2008-08-30 18:08
Thread for non-PrimeNet LL testing	ThomRuley	Lone Mersenne Hunters	6	2005-10-16 20:11

2006-02-10, 03:49	#23
Yxine "Aleksandr" Nov 2005 Russia 90₁₀ Posts	I have 5+ computers available and like to participate in beta-testing!

2006-02-13, 09:20	#26
Old man PrimeNet Jan 2003 Altitude>12,500 MSL 101₁₀ Posts	Two years ago this design document initially captured the v5 system concept and objectives, though it's now woefully outdated and lacks details for feature-level implementation. However, the v5 GIMPS client web API specification is current for the system in development and describes data-level dependencies that hint at how the system works and what will be available for users on the web site. A single web interface (a set of forms/pages) will manage all the computers of an account, as a whole or specific computers, and its operating options and preferences. Another set will manage all the resources of a team, which is created and managed by a teamleader. An important design philosophy of v5 is allowing users to manage distributed computing resources across multiple projects and on various workloads without having to physically revisit their computers once they have installed the v5 client. Before you can install your first v5 client program, you normally need to create an account on the web site. This sets up an account summary page you can optionally make public and to which you can link or frame-embed from elsewhere, and enables your access to manage your account, computers and workload. It also provides an account binding the v5 client expects when you install it, though it's also possible to install a v5 client with an anonymous account binding that is instead managed by the project admin (in GIMPS' case, that's big GW). When you install a v5 client, you give it just your account name, an optional computer name, and nothing else! This can be setup for batch installs, too. Once installed the v5 client does not need to keep a record of who owns it or its name, just its identity key. From that point forward the v5 server tells the client, when it asks, who owns it and what its name is. Installed v5 clients will appear on your personal admin web page. You can still manage the client directly at the computer to some degree (including changing its account owner and local options and work preferences), but for the vast majority of the time you will select your project and work-type preference options on the web and your group (or cloud) of computers will simply perform your wishes on their own, collectively, or even individually or even as different CPUs on a computer if you want to fine-tune them. You will also be able to rename computers and transfer them between accounts. The local v5 client settings, however, will always have the ability to over-ride your web account (or team) settings. You choose your optional team on your admin web page and can change it as often as you wish - your computers ride along with you. We haven't finalized decisions about team-level computer control, but we are carefully thinking it through. The current plan for crediting work in teams is that whatever team-level credit you earn while you are part of a team stays with the team. The same plan applies to computers on accounts - you can give away or lose the computer but the work credit it earned for your account stays on your account. Much of the alpha & beta test phases will be focused upon these web management tasks, teams, and stats reporting rather than the v5 client and server. For alpha in particular we don't need a ton of computers per person - only enough to work out the kinks until beta. At first the v5 web site will be a little rough around the edges until the functionality is firm, but we'll pretty it up later. Get it working, then tune it up.

2006-02-21, 16:20	#29
Old man PrimeNet Jan 2003 Altitude>12,500 MSL 101 Posts	PrimeNet v5 will limit assignments to a single computer to some rational threshold based upon its speed, work history and results reliability because the amount of work a computer can per unit time can be very accurately determined as a function of the exponent. This will probably make using a single computer as a front-end for others off-net impractical, though undoubtedly some will try anyway. Others may study the PrimeNet 5.0 GIMPS APIs and write scripts to emulate a client program, however the security features will treat these as open-source untrusted clients, which are subject to tighter safeguards and risk being locked out completely in the event of counterproductive activity. They also are metered to rational rate limits. An important objective of the new system is to enable users to manage their group of computers from the server's web site and within the wider guidelines of a project's general goals, rather than resorting to sneakernetting data from computer to computer. The server will be much more selective about what work to give computers to achieve the best results within the user's configured options, so there's less reason to invent strategies to get better assignments. Overall, we would rather add any needed fine-tuning workflow features to the v5 server than encourage script-writing and sneakernetting to achieve the same effective result outside of the server's workflow visibility. Defeating PrimeNet's assignment system for personal reasons (climbing stats, etc.) will be less efficient for the project, if not simply counterproductive. Ideally, we want every computer participating to run a v5 client on the Internet - but we also realize that's not going to work for a few folks with special network issues or client programs without PrimeNet support. Bottom line: the easiest way to successful participation will be to use a trusted v5 client program or an open-source untrusted v5 client code. delta_t: The reports are likely to change somewhat, but at a minimum (early on) the same information will be presented. We plan to add several new statistical reports but this will likely be among the last progressive stages of the upgrade.

2006-02-21, 20:39	#32
gribozavr Mar 2005 Internet; Ukraine, Kiev 11×37 Posts	fact_trial_LMH -- what range of exponets will be factored? Will 100mdpp fall into this? Since there is LL_100M, why not add fact_trial_100M to avoid confusion?

2006-02-21, 20:47	#33
tha Dec 2002 5·163 Posts	Great, how about: LL_10M_check