How to restructure multiple users from V7 to V8

[appepi]

When I first started folding in 2020, mostly using Windows 10 on HP Z- series workstations, I called myself a "Team" and each Machine was a "User", with its own passkey etc. This works fine. The machine/user-specific FaH stats and EOC stats make it easy to check if each device is working normally and so on. However, I had to re-think this for 8.4.9 and I decided to try to get by with just one account, mainly to avoid the nuisance of multiple 4 word pass phrases when I fire up a device. And no, I can't just let them pause because they still burn power when doing zip.

OK, so I created an account for 8.4.9 as "z441u" so that's just one pass phrase that eventually I might be able to remember, and I connected this Machine (Z605 aka "User" Z605 in the old scheme) and has its own passkey and is eligible for bonus etc. Z605 is the device I usually use and would be the only case that isn't 99% for folding only. It has 2x Xeon X5660 (6X 12T) and an antique Quadro 600.

So, I uninstall 7.6.21 and install 8.4.9. All goes well, it connects, it knows that Z605 contributes to Team appepi. So far so good. Of course it grabs all 24 "CPUs", but because it's a HP workstation I can use the HP Performance manager and assign CPU affinity to Core A8 so it can only run on 12 even-numbered threads - equivalent to switching off hyperthreading. The job runs these up to 100% use, the maximum core temps are higher than I would really like (80C vs 75C) but the average temp is fine. I note that the PPD figures assume the bonus so I think that's nice too but not too surprising since 8.4.9 knew about the Team so presumably knows the Bonus status of Z605 also. Later I find the slider that would let me say that FaH can only have 12 cores, but I figure to test that with the next device. I also check that there's no obvious way for me to use the "expert" slot settings in 7.6.21 that let my RTX2060s avoid project 18251, so I figure those 3 devices can stay on 7.6.21.

Moving on, I add Z805, which only has a single X5620 (4C 8T) but also a GPU (GTX 1080). I note that 8.4.9 reserves a whole Core (ie 2 logical CPUs) for the GPU leaving 6, which it gives to a CPU job. I note that temperatures remain low so I don't bother to restrict further. The GPU starts OK after I notice the "enable" button and is running with the PPD I expect so clearly 8.4.9 knows that Z805 is eligible for bonuses. Then I climb up to the attic and add Z803 to this wonderful new World of 8.4.9. Z803 also has a GTX 1080 but has 2x Xeon X5675 (6C 12T) so I use this setup to see what happens if I just tell FAH it can only have 12 CPUs for the CPU client, by watching the utilisation of each thread in HP Performance manager. It seems to grab them all from CPU0, so I use performance manager to spread the Core A8 load across CPU0 and CPU1, while saving a full (1C 2T) for the GPU. (or rather for Core_26).

All of this takes a while, and some jobs finish. Now I can find out how they are attributed. Shock/Horror but not total surprise. They are all attributed to user z441u, and since user z441u is not (yet) eligible for bonuses, I don't get them. No doubt this bonus will be a temporary problem until more completed WUs have been done, but if 8.4.9 knows that the "machines" are users eligible for Bonuses (when reporting the notional PPD), why does it credit the points to the Fahcontrol account z441u and refuse bonuses? There seems to be some mixed communication happening here.

Anyway, is there a way of ensuring that a particular "machine" is associated with a particular user (and its bonus and team association) rather than being absorbed into the "account?" Or am I going to have to learn umpteen passphrases after all? I realise that 8.4.9 is designed for power users with umpteen 5090s to manage, rather that curators of PC museums, but still ....

[calxalot]

If you link a client to an account, the user, team,passkey, cause of that account is periodically synced to all linked clients.

[calxalot]

The client does not know if a user +passkey is eligible for bonus points. It simply assumes the bonus is active when a passkey is set.

[calxalot]

If you want different users for each machine, you cannot link them to the same account.

[appepi]

Thanks to calxalot for the clarification.

If you want different users for each machine, you cannot link them to the same account.

But after linking them all to the same account and seeing how that goes, my final conclusion was that remote management is good, but an inflexible implementation that requires all the "Machines" to be drones serving the Account Master reminds me too much of that electric car company that used to exist in the USA - though perhaps it wasn't a great idea to watch Abandoned Americana while I was doing this.

Also 8.4.9 is very sticky in defaulting to whatever best suits endless maximum-consumption folding. That is REALLY annoying.

So I decided to de-link them, and that proved tricky. In 8.4.9 they really WANT to be slaves. Like, how does a "Machine" know what "User" to credit with its work unless someone tells it? This does not demand that they should all be stuck with whatever user is specified by the Account, which persists because as far as I can tell there is no way for a de-linked machine to be told to send its work anywhere other than the user it was last told by the Account.

So I decided to set up multiple accounts. No problem at one level: all my diverse "Machines" are equal in having their own Username which is the same as that of the Windows Machine, and each has its own Passkey so they can all be included in the Fold. So I try to create a new account for each of them too. No problem until I try to save it, when I am told it already exists. After some further fun and games my conclusion was that the unique key to the Fah system "Account" is just the email, of which I don't have enough to go around.

Thus the alternative formulation seems to be:

If you want different accounts for each machine, you cannot link them to the same email.

So now all I have to do is go back to where I started on Thursday, namely good old 7.6.21, though along the way I might make a stopover at uninstalling and reinstalling 8.4.9 to see if I can tell it what user to credit.

[appepi]

along the way I might make a stopover at uninstalling and reinstalling 8.4.9 to see if I can tell it what user to credit.

Well, that was a good idea, though I say so myself as shouldn't. After uninstall/ reinstall, the 8.4.9 on this Z605 machine had nothing prefilled by the Account in its settings, so I typed in the ones for Z605 and off we went. Then I climbed up to the attic and simply edited the others (Z601, Z602, Z604, Z802, Z803) to assign them as I had previously done with 7.6.21. I assume that all of this would be undone by the dreaded Account if I linked them because (say) I simply wanted to watch what was going on without climbing all those stairs, so my basic point remains: I don't see the necessity for Management to insist that they all serve the same User, but at least 8.4.9 is willing to let Machines flourish as individuals as long as they don't talk to one another. It could be worse.

[muziqaz]

Why do you want separate users for your each machine?
v8 has WU history section which stores last 500 WUs received by each of the machines. It can be filtered by machine name (which can be changed to your heart's content). Then you can split your devices within each computer to resource groups, which can also be named whatever you like.
That alone is more than enough to see how each computer is doing

[appepi]

Fair points. The Forum died while I was editing a reply, but just as well because I was planning to trial 8.4.9 further at that stage. However, I had set some devices in the attic to "Finish" because I was finished testing them. Team Appepi's chart on Extreme overclocking shows a giant leap to around 10M PPD as a result of having every eligible device doing GPU or GPU+CPU folding for 24 hrs/day. Unfortunately my electricity use chart is just as extreme, so it was time to calm down. Thus I was furious that "Finish" in 8.4.9 does not seem to be the same as "Finish and Stop" in 7.6.21. Yes, the WU had finished, but then the machine had picked up a new WU. Then I thought maybe the ongoing GPU job on Z803 was (via Claus 23X subpar Vii of the Fah Forever Act) sufficient to tell the Machine not to stop, and made made the Machine pick up another CPU WU. So I killed the CPU and waited out the ETA on the GPU - WHICH PICKED UP ANOTHER GPU JOB. And another device that was only running a CPU job did the same. That is a total deal-breaker, so irrespective of any other virtues of 8.4.9, the devices with eligible GPUs are being converted back to 7.6.21 so I can use the LARS timer to control the Finish and Stop function. So either 8.4.9 gets a proper WU-specific "Finish and Stop and don't start another WU of this type (GPU/CPU) on this machine until further notice from your local human being" or 7.6.21 and I will have to finish up and stop folding together.

[muziqaz]

Finish on v8 is exactly the same as finish on v7. It works as intended. I have no clue why your finishing devices on v8 are picking up new WUs

Edit:

Here: https://www.youtube.com/watch?v=cYV8xRr7F-Y

[calxalot]

Not quite the same. On v7, a slot would finish and stop, but would start folding again the next time the client was launched.

On v8, the pause after finishing is sticky. You describe a mystery bug where finish does not pause. Such has never happened to me.

[muziqaz]

Not quite the same. On v7, a slot would finish and stop, but would start folding again the next time the client was launched.

On v8, the pause after finishing is sticky. You describe a mystery bug where finish does not pause. Such has never happened to me.

v8 finish is still finish, and it works. And yes, I have never seen finish to not work. Or at least I don't remember

[arisu]

I think the web client sends a "run" command and then a "finish" command a second later when you click finish. I don't think a bug can be triggered there but if you directly send finish over the websocket (such as with fahctl or lufah) and it happens immediately after another command, it might not get picked up.

[appepi]

Thanks for the info that 8.4.9 "Finish" should behave in the same way as 7.6.21 "Finish and Stop". All I can offer in reply is that it didn't on a device (Z604-CPU only ) left running overnight to finish and stop a WU that was running with ETA 3am, as per a note I left on it. But the WU history showed it had completed the old WU and taken on a new one. I dumped the WU, uninstalled 8.4.9 and reinstalled 7.6.21 and am running one WU with expensive peak electricity to make up for the dumping.

On the other hand, Z805 seemed to have behaved properly and was idling. It is the first one I checked before I climbed up here to the attic to check the others, so it is still on 8.4.9. I will keep it as a test bed. And maybe re-visit the issue when 7.6.21 stops doing everything I need done.

[calxalot]

There have also been issues if the system clocks are off.
If it’s far off, web control may not connect to servers.

[appepi]

I'm pretty sure that all my Windows devices are set to synchronize automatically with the time server time.windows.com, but I'll check next time I visit the attic.

There have also been issues if the system clocks are off. If it’s far off, web control may not connect to servers.

Though synchronicity might be a problem since, according to EoC, Team appepi was passing through a major gravitational anomaly at the time.

674 Team Romania
675 appepi
676 choderalab
677 iridium

PS: I now have Z601 (CPU only), Z805 (GPU only), and this PC Z605 (CPU only) running WU's under de-linked 8.4.9 and set to "Finish", so we'll see how they go.