LHC@home

... I'm running virtual box 7.1.4 ...

This version runs fine.

The reason why it sometimes fails is a race condition when you had no ATLAS tasks and then start at least 2 of them concurrently.
If you are lucky the timings do not cause the race condition and everything works fine.
Otherwise the media registry gets corrupted and stays corrupted until you manually clean it.

Vboxwrapper 26208 includes a patch that avoids the race condition.
CMS/Theory use a beta version that already includes that patch:
https://github.com/BOINC/boinc/pull/5571

Has anyone else been able to run more than one atlas task at a time? The instant I do I end up with the yellow hard drive triangle in virtual box and all my other atlas tasks fail with computation errors until I clean it up.

on several of my hosts I run more than one Atlas task at a time. So no idea what exactly might be the problem on your host, but it looks like some misconfiguration of the VirtualBox. Which version are you running? Maybe an update to a newer one might help.

How much more ram do you think they need? I'm running 1024 mb as suggested in the thread.
Do we think maybe 1280 mb might be enough?

With the amount of RAM you have, I would suggest 1536MB

On one the the two hosts which are currently running Herwig7, I set the RAM to 1024MB. Console_3 shows about 48MB swapping.
On the other hosts on which 13 Herwigs are currently running, I set the RAM to 1536MB. Within the first 4-5 days' runtime, there was no swapping. I now checked again and see values between 4 and 18MB. All tasks are still in the integration phase. I am courious how swapping will be thereafter. So for future Herwigs, I might end up with even a higher RAM than 1536MB (with 128GB I've plenty of RAM available, anyway).

You think I should just reset lhc at home all of mine are still failing. I do have a squid cache, so maybe that's the source of my issues?

ATLAS EVNT files are very large and Squid does not cache them.
The decision which file needs to be downloaded is made by the BOINC client based on the WU data it gets from the server.

Hence, I don't think your Squid is the problem.
Nonetheless, feel free to purge it's cache - it will rebuild automatically.

A project reset looks more promising (no 100% guarantee).
Finish all tasks in your work buffer before.

Yeah the squid purge didn't work I'll set no new tasks and go for the project reset.

You think I should just reset lhc at home all of mine are still failing. I do have a squid cache, so maybe that's the source of my issues?

ATLAS EVNT files are very large and Squid does not cache them.
The decision which file needs to be downloaded is made by the BOINC client based on the WU data it gets from the server.

Hence, I don't think your Squid is the problem.
Nonetheless, feel free to purge it's cache - it will rebuild automatically.

A project reset looks more promising (no 100% guarantee).
Finish all tasks in your work buffer before.

2023-02-07 15:05:56 (46556): Adding network bandwidth throttle group to VM. (Defaulting to 1024GB)
.
.
.
2023-02-07 15:06:02 (46556): Setting network throttle for VM. (5120KB)

It looks like you tweak your network bandwidth settings for the VMs (or BOINC as a whole).
This makes no sense since it applies only to outgoing traffic (from the VM perspective), but it may affect the connection timing.

You may leave those settings unlimited or at default values.