| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Beta Testing
Forum: Beta Test Support Forum
Thread: Norton Insight cancelled all my beta10 autodock 6.07s |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 31
|
|
| Author |
|
|
coolstream
Senior Cruncher SCOTLAND Joined: Nov 8, 2005 Post Count: 475 Status: Offline Project Badges: 
|
That's a good point.
----------------------------------------Fingers crossed that the next betas come soon to test your theory and my configuration   Crunching in memory of my Mum PEGGY, cousin ROPPA and Aunt AUDREY. |
||
|
|
rembertw
Senior Cruncher Belgium Joined: Nov 21, 2005 Post Count: 275 Status: Offline Project Badges: 
|
...will they no longer be counted as being reliable;.. As far as I know, every computer has a "reliability" and "turn over" variable stored somewhere in WCG. Again, as far as I know, your rating may have worsened because of the errored Beta's although I assume that WCG doesn't take those into account. If WCG does, then they ought to change that. Still, it is a variable and not a constant meaning that returning normal work units errorfree will improve the status again. Maybe the mods or techs can elaborate on this, or correct me if I'm wrong, but the end of the story, in short, is: don't worry. Everything will correct itself after some time. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The reliability rating has nothing to do with beta eligibility, is used to determine which PCs can process the rush jobs for WUs canceled, errored, or aborted.
|
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Those who've paid attention will have seen someone reporting with a 6 project profile and getting for all six a 'no work ...' Quota 6 used. i.e. each day the client if totally unreliable is tested for all selected projects if a problem was resolved... after all it would be silly if a client always fails on for instance HPF2, but not on other projects, to not send work if the project selections permit. If one or more projects fail the re-test, they're being bounced to try again until the 00:00 UTC hour has passed.
----------------------------------------Now unless the quota is set to 1 per day due persistent failure, the client is not going to ask for work at all until midnight [UTC]... Now which beta launches at that hour ;? Full reliability on fast machines is quickly restored. Think the number is 15 tasks without error... the restore rate is exponential going from 1 > 2 > 4 > 8 > 16 > 32 etc, where the daily top out per device is somewhere in the ~320 range per day. Sometimes manually upped such as was done when there were many many shorties. Also manually set down to 1, if a farm goes bunkers. Not sure if WCG allows for all cores to receive a reassessment task. Probably it will want to test 1 before granting more once a host has received the red card. 2 cent end
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
The reliability rating has nothing to do with beta eligibility, is used to determine which PCs can process the rush jobs for WUs canceled, errored, or aborted. While a computers error-rate doesn't matter when it comes to being assigned beta-jobs, if beta-jobs erroring-out has an influence on reliability-rating it does matter for all the non-beta-work computer is doing... Taking a look on the relevant server-code, if WCG haven't made some special changes, the rules are: 1: All new computers starts with error_rate = 0.1 2: In Validator, each success multiplies error_rate with 0.95. Each validation-error increases error_rate with 0.1, max 1. So, interestingly, any download-errors and so on won't change error_rate, only "success"-tasks failing or succeeding validation will change error_rate. But, while error_rate isn't changed, Scheduling-server uses these rules when deciding to send "reliable" work or not: 3: If host.error_rate > project-limit => not reliable. 4: If host.avg_turnaround > project-limit => not reliable. 5: If host.max_results_day < config.daily_result_quota => not reliable. So, any beta-tasks erroring-out will lead to computer becoming unreliable, since the quota will decrease. But, the quota is doubled when next "success"-task is reported. Meaning, even with a long string of failing beta-tasks, computer will normally be reliable again after a single "success"-report. Due to the very long run-times on the DDDT-beta's, it's maybe also possible the average turnaround-time gets too long, especially if misses the deadline since this immediately increases avg_turnaround to how many days the deadline was, so can lead to misses the reliable-rating. But, then host starts getting valid tasks again, the turnaround-time will quickly drop again, since avg_turnaround = avg_turnaround * 0.7 + new_turnaround * 0.3 (since validator calculates this, it can still take some time if needs replication to pass validation, for avg_turnaround to start dropping again). A beta-task that somehow is reported as "success" but later fails validation on the other hand, can have a much bigger impact. I've no idea what WCG has set the limit at for being reliable, but if it's example 0.01, one task failing validation will need 47 tasks passing validation before becoming reliable again. Worse-case, if a string of errors brings error_rate = 1, you'll need 90 tasks passing validation to become reliable again. So, beta-task-errors does influence the reliability-rating, but as long as they're reported as errors it's only influencing the quota, and a single "success"-report will normally bring quota back to max so is reliable again. Even worse-case with a quota of 1, after 7 "success"-reports the quota will be 128 per core, this is AFAIK larger than WCG's daily quota per core... Note, user-abortions is also counted as errors. The much less frequent instance of beta-tasks failing validation will have a much larger impact, before host becomes reliable again... edit - sees Sekerob has made a post in the mean-time, only going on quota and not the other points, so no reason to change anything (appart for a few typing-errro).  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." [Edit 4 times, last edit by Ingleside at Oct 11, 2009 12:39:38 PM] |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Those who've paid attention will have seen someone reporting with a 6 project profile and getting for all six a 'no work ...' Quota 6 used. i.e. each day the client if totally unreliable is tested for all selected projects if a problem was resolved... after all it would be silly if a client always fails on for instance HPF2, but not on other projects, to not send work if the project selections permit. If one or more projects fail the re-test, they're being bounced to try again until the 00:00 UTC hour has passed. If WCG haven't made various extensions to the server-code, all tests is done per host, not per sub-project. Meaning, an unreliable host will be unreliable in all sub-projects, not only example HPF2. BTW, WCG for some reason has changed the quota deferral, so if you fails with "reached daily quota", you'll get a 4-hour-deferral, and not until "midnight server-time + random 1 hour" as is normal... Also manually set down to 1, if a farm goes bunkers. Not sure if WCG allows for all cores to receive a reassessment task. Probably it will want to test 1 before granting more once a host has received the red card. The quota saved in database is per core, but is multiplied with #cores when schduling-server tests for quota. Meaning, even with a quota of 1 an 8-way system will get 8 tasks. Well, projects can limit how many cores should be counted max, so example a 128-way system maybe still only gets 8 or something... "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Ingleside, you've earned yourself an 'interesting', theory. Think knreed wrote in about 1 week ago on the exact work check order in a way the general membership will have understood. There's lots of server code that needs/is to be 'contributed' back to Berkeley, so don't assume on server settings unless you know from extensive observation... the Dr. did not want to receive the code until the server software version at WCG were aligned with Berkeley was my off-line interpretation. Now what do you think that means?
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Ingleside, you've earned yourself an 'interesting', theory. Think knreed wrote in about 1 week ago on the exact work check order in a way the general membership will have understood. There's lots of server code that needs/is to be 'contributed' back to Berkeley, so don't assume on server settings unless you know from extensive observation... the Dr. did not want to receive the code until the server software version at WCG were aligned with Berkeley was my off-line interpretation. Now what do you think that means? It's been posted before that code to detect a single project (like proteome folding) not working on a host won't be started coded before next year. Therefore I find it unlikely there's already code that tracks error_rate and quota for different sub-projects independently, since if this was the case it's enough to add a couple lines to the scheduling-server to stop users downloading a bunch of wu in a sub-project that will only error-out... Since there's no indication when a host is reliable or not, can only guess, and as for the error-limit, it can just as easily be set to 0.05 or something like 0.01.  The only you can really test is how the quota responds, by aborting a crapload of tasks...  As for adding code-changes back to BOINC, it's mentioned 1st. June https://secure.worldcommunitygrid.org/forums/...ead,25792_offset,0#232347 but WCG still haven't upgraded their scheduling-server so... "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." [Edit 1 times, last edit by Ingleside at Oct 11, 2009 3:57:10 PM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
On AV, I'm really pleased to have updated from the free Alwil Avast 4.7 to the 5.0 Beta 2. An appealing and expanded interface, exclusions, custom scan profiles, even a time limited suspend to do certain install / updates that don't like active AV programs... 10 minutes, and it kicks back in without being forgotten. The bonus today, a dormant program who's language files had a suspect string, moved to the security chest for verification asking for approval. BOINC data dir set to the exclusion zone, since it's already in a sandboxed zone :>)
----------------------------------------A feature that ZA users might be jealous off: Ignore local communication - if this is checked, avast! will not scan communication between applications running on your computer. Unchecking this box will result in local communications being scanned, which may slow down your computer slightly. Meaning, GUI to Core client comms is ignored, a frequent issue when the BOINC Manager tries connecting to the service.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Oct 19, 2009 12:08:23 PM] |
||
|
|
sk..
Master Cruncher http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif Joined: Mar 22, 2007 Post Count: 2324 Status: Offline Project Badges:
|
Most Antivirus suites now contain software designed to spot new, never seen before, threats! So false positives are sometimes generated, and frankly, cause more problems than many would be threats.
----------------------------------------For most people, live continuous scanning of every file you open, is overkill. Most people dont have email servers, sensitive or confidential data on their systems. So if you do a regular backup you donât have that much to lose. System functionality is at least as important as security! A reasonable antivirus security setting would be to scan the computer once or twice a week and you should really stop everything else running when you do this. It is more important to clean the cache or do a backup than scan every process continuously. I use Spybot Search & Destroy on systems because it has a block list of websites, preventing you from opening a site that is known to contain malware. AVG is also an excellent choice for a free antivirus scanner, and there are many other simple, free programs that do a good job. I try to avoid the overburdening security suites. They are overtly invasive to the extent they slow your system down, sometime to the point that you cant use it, and most have at one time or another frozen just about every kind of operating system they have run on. They are also time consuming to maintain, run, upgrade and can be quite costly. If you just do common things such as browse a few internet forums, look up the news or weather, check the sports results, send a few emails and run WCG, you donât have too much to worry about, especially if your built in firewall is switched on or your router is using some security features, you have the basic sense to use a password on your system and use Windows Defender. [Edit 1 times, last edit by skgiven at Oct 24, 2009 1:44:06 PM] |
||
|
|
|