Reduce your cache size so you aren't overcommitted. At least until the DCF for the project comes down to the realities of this new mix of Einstein work.
Sorry the flops estimation was still off by a factor of ~5. Workunit generated from now on should have the correct one. This should also affect credit, a task from a recently generated workunit should get you 10k.
Sorry the flops estimation was still off by a factor of ~5. Workunit generated from now on should have the correct one. This should also affect credit, a task from a recently generated workunit should get you 10k.
I happened to be working on a test machine at the time you made the x5 increase. I had a small cache of work, initially estimated to take ~23 mins each, and the first of those had finished, taking a little over 2 hours and thereby setting all others to that same adjusted time estimate. From other test hosts running, I expected that ~2+hrs would be very close to correct.
At that point I decided to get a few more tasks and finish up for the night. I made a work request expecting to get about 5-10 more tasks and I got a single task with an estimate of over 10 hours. It took a while to get over the shock but I did eventually look at the ratio between a former task estimate and the new one and found it to be precisely 5, almost to the very second. So I guessed what you had done and the mystery was solved.
At that point, there was no message from you about this so I started to compose a message in this thread to warn others about some of the consequences of this dramatic change. I had been keeping up with any posts about the new Bu tasks and was definitely logged in earlier in the day. I spent a lot of time composing a fairly detailed post about what to expect, especially when resends arrived for tasks that predated the 5x change. When I tried to preview what I had written, I got a message that I was not authorised to post in this forum and what I had spent more than an hour composing was gone. It was late so I quit for the day.
If you need to have a system with such short logins, could you please implement a warning message, perhaps the moment a person tries to compose a new message or a response, that they need to renew their login before proceeding? It's currently very easy to NOT notice you've been logged out at some point and this is very frustrating if you're trying to do several things at the same time.
Regarding future resends for tasks created prior to the above 5x change, they will show an estimate 5x smaller than what they will eventually take when processed. This is going to create havoc for some people when trying to manage the size of the work cache. How difficult would it be to make a 5x adjustment to the flops estimate just for those resends of tasks initially created with the former lower value? Presumably, the code that creates a resend could check and adjust the flops value if needed?
Sorry for that, in particular the delay of the post. It was a busy day with a couple of different problems that I needed to resolve more or less in parallel.
I just fixed the fpops estimate of the already generated workunits from which resent tasks are instantiated, so these should all be the same now.
I'm really sorry to hear about your session expiring, this can be really annoying. It hit me on other websites, which allow only a few minutes to submit a single form. I'll see what I can do about the session duration.
What I usually do for posts that I expect to be longer is to prepare these in an editor first and then just paste it in the browser. Occasionally I also post a quick and dirty version, then edit it a lot. This may also work in a post that you (as a moderator) can hide immediately, and unhide when you're done. Recently I also developed the habit to just mark all text and copy it to the clipboard before hitting any send or post button, just to have the text somewhere in case something goes wrong.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I've seen the same thing but
)
I've seen the same thing but not so dramatic.
Reduce your cache size so you
)
Reduce your cache size so you aren't overcommitted. At least until the DCF for the project comes down to the realities of this new mix of Einstein work.
Sorry the flops estimation
)
Sorry the flops estimation was still off by a factor of ~5. Workunit generated from now on should have the correct one. This should also affect credit, a task from a recently generated workunit should get you 10k.
BM
The recent adjustment of the
)
The recent adjustment of the flops estimate is meant to fix that. Sorry it took us so long. We may still need to tune it.
BM
Bernd Machenschalk
)
I happened to be working on a test machine at the time you made the x5 increase. I had a small cache of work, initially estimated to take ~23 mins each, and the first of those had finished, taking a little over 2 hours and thereby setting all others to that same adjusted time estimate. From other test hosts running, I expected that ~2+hrs would be very close to correct.
At that point I decided to get a few more tasks and finish up for the night. I made a work request expecting to get about 5-10 more tasks and I got a single task with an estimate of over 10 hours. It took a while to get over the shock but I did eventually look at the ratio between a former task estimate and the new one and found it to be precisely 5, almost to the very second. So I guessed what you had done and the mystery was solved.
At that point, there was no message from you about this so I started to compose a message in this thread to warn others about some of the consequences of this dramatic change. I had been keeping up with any posts about the new Bu tasks and was definitely logged in earlier in the day. I spent a lot of time composing a fairly detailed post about what to expect, especially when resends arrived for tasks that predated the 5x change. When I tried to preview what I had written, I got a message that I was not authorised to post in this forum and what I had spent more than an hour composing was gone. It was late so I quit for the day.
If you need to have a system with such short logins, could you please implement a warning message, perhaps the moment a person tries to compose a new message or a response, that they need to renew their login before proceeding? It's currently very easy to NOT notice you've been logged out at some point and this is very frustrating if you're trying to do several things at the same time.
Regarding future resends for tasks created prior to the above 5x change, they will show an estimate 5x smaller than what they will eventually take when processed. This is going to create havoc for some people when trying to manage the size of the work cache. How difficult would it be to make a 5x adjustment to the flops estimate just for those resends of tasks initially created with the former lower value? Presumably, the code that creates a resend could check and adjust the flops value if needed?
Cheers,
Gary.
Sorry for that, in particular
)
Sorry for that, in particular the delay of the post. It was a busy day with a couple of different problems that I needed to resolve more or less in parallel.
I just fixed the fpops estimate of the already generated workunits from which resent tasks are instantiated, so these should all be the same now.
I'm really sorry to hear about your session expiring, this can be really annoying. It hit me on other websites, which allow only a few minutes to submit a single form. I'll see what I can do about the session duration.
What I usually do for posts that I expect to be longer is to prepare these in an editor first and then just paste it in the browser. Occasionally I also post a quick and dirty version, then edit it a lot. This may also work in a post that you (as a moderator) can hide immediately, and unhide when you're done. Recently I also developed the habit to just mark all text and copy it to the clipboard before hitting any send or post button, just to have the text somewhere in case something goes wrong.
BM
O3ASBu search
)
O3ASBu search progress
This is the Server Status ftm.
maeax wrote: O3ASBu search
)
https://einsteinathome.org/server_status.php
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Yep, this happens for 1d when
)
Yep, this happens for 1d when I add a chunk of new work. Should fix itself around midnight (UTC).
BM
I started getting these again
)
I started getting these again this morning