| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: No need to validate the validator? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 6
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I may have this wrong, but I think I spotted a way to (albeit slightly) reduce the number of work-units that are sent out.
Consider this scenario for a zero-redundancy project: WU_0 is sent to Machine_A. Machine_A requires validation for this project, either because it is unreliable or for a random check, so a second copy, WU_1, is send to Machine_B. I'm assuming that Machine_B is required to be a known reliable machine -- if this is not the case then what I'm about to suggest doesn't work! What happens if WU_0 errors out, or is too late or whatever? At the moment it seems to me, from looking at the logs for my machine, that a third copy, WU_2, is sent out to Machine_C, and the WU is validated between Machine_B and Machine_C. But WHY?!? IF the ONLY reason WU_1 was sent out was to validate Machine_A, and Machine_B is a known reliable machine, then the result from Machine_B can be trusted and there is no reason to send out WU_2 at all (unless there's problem with Machine_B, of course). It may be that keeping track of this situation causes more load on the server than it's worth, but I'm thinking that the time saved by not sending out WU_2, indeed the fewer WUs sent out altogether, the quicker the science is completed. |
||
|
|
Mamajuanauk
Master Cruncher United Kingdom Joined: Dec 15, 2012 Post Count: 1900 Status: Offline Project Badges: 
|
Apis_... I tend to agree
----------------------------------------However, does this retain the integrity of the results in the same way that the current system does? If the trusted machine, has a problem and it's results are not checked, then the results are at best questionable, at worst unreliable/wrong/inaccurate! End result potentially a complete waste of crunching time. While the idea is sound, it is easy to see why the results are checked in the way they are. Does anyone else agree?
Mamajuanauk is the Name! Crunching is the Game!
  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Reliable hosts have to be verified too as noted and yes at times when there's something wrong with _0, two extra copies are send out at the same time. My stats suggest that compared to the server 601 revalidations, the number went from 20% to about 25% with server 700. That would be a 25% increase in extra copies. Think this was a Berkeley "out of the box" effect, not by WCG tinkering design, but the only one that could comment on that is WCG techs. IIRC they're aware of this, certainly mentioned this to them.
----------------------------------------FAIK, if a result is "No Reply" (aka Too Late, but only marked Too Late is still crunched and returned), only one extra copy is send out. In case of Invalid, I'd say 2 need to confirm as something could be awry with the task assembly or the application itself. I'd rather err on the [Edit 1 times, last edit by Former Member at May 2, 2013 5:44:06 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
@SekeRob
I agree about erring on the safe side, but I was just floating something for the techs to consider in the case that no-one had spotted it as a possibility. I'm just going to step back and make no further comment because I simply do not have enough knowledge. |
||
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
Apis,
Your suggestion is interesting, but if I'm remembering correctly the second result sent on the initial test does not require a reliable host. It is a test to make sure two different hosts still get the same data. On the backend server, adding the ability to check if one of the hosts is a reliable host would add extra stress on the servers for a minority of the cases. I do not think this added stress would be much, but doing a little extra validation never hurts :) Please if you have other suggestions or ideas we are happy to hear them. Thanks, -Uplinger |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
... the second result sent on the initial test does not require a reliable host. OK, so I'm off base. Never mind, and thanks for taking my suggestion seriously. And do not fear. I'm not backward at making suggestions in the rare case that I spot something that I think might help.  |
||
|
|
|