2022 Queue management of 15e

[RichD]

Originally Posted by swellman

Also the new job file does not seem to be producing results.

Just a speculation but all the unsent queued WUs need to be sent before the "new" WUs will be sent out.

[VBCurtis]

The new job should be about half the Q-range of the old one to get just the additional rels needed. As set now, you're gonna be like 40% oversieved. I didn't test-sieve, but I bet 35-125M will be plenty when combined with the old relations.

[kruoli]

Originally Posted by swellman

Greg - if you can please hold off deleting anything until Oliver and I can download the old relations file.

That should not take long. It says there are 2.7M relations still underway for the old job. Should I wait some time to have a majority of those WUs tickle in? Or does that delay something?

The new job has started getting relations, by the way.

[swellman]

Originally Posted by kruoli

That should not take long. It says there are 2.7M relations still underway for the old job. Should I wait some time to have a majority of those WUs tickle in? Or does that delay something?

You can download them whenever you like. I’d recommend sooner rather than later, just so you get them. The now 1.8M remaining relations won’t make that much difference in the big picture.

The new job has started getting relations, by the way.

So I see. Greg must have intervened. Glad the WUs from the -a side are flowing now. Once it finishes, just download and combine with the data file from the -r side. As noted by @VBCurtis, there will be a LOT of extra rels but those will disappear with remdups or in filtering within msieve.

If you come up short on rels I can bump up Q as needed on the -a job of course.

[kruoli]

For the record, I have a copy of the old job now. So if it is gone, we will be fine.

By my original estimations, 35M-125M should give a bit more than 260M raw relations, so around 520M in total. I am not experienced in how the duplication rate will look like in this case (partly -r and party -a), but when this would not be sufficient, we have a great starting point; thank you!

[chris2be8]

I did a job sieved on both sides on 14e, I found I had to sieve about 50% more special-Qs in total than sieving on one side would have needed. Which confirmed that sieving on both sides was not useful unless we were limited by what siever we could use.

It was 29^173-1, jobnames f29_173m1a and f29_173m1r. Search for posts relating to it for more details.

For 145!-1 the speed up from the algebraic side will probably beat the higher duplicate rate, but I'm not sure exactly how many relations you will need.

[kruoli]

There is only a handful of WUs available; we should move back at least one of the other jobs.

[swellman]

Originally Posted by kruoli

There is only a handful ob WUs available; we should move back at least one of the other jobs.

But are there enough rels now on the -a side of 145!-1_b to build a matrix once combined with those from the -r side? It will probably be several days until we will know with any certainty, and I agree it’s time to restart sieving on another job, but I’m just suggesting we might have to bump up Q on 145!-1_b and sieve it some more in the near future. Hoping not but…

[kruoli]

If we go by the numbers of Chris, we will need more relations, but I am willing to wait for our current WUs complete up to 10M outstanding, then I will download them and will give remdups4 a go to determine how much duplicates there are.
If you want to bump it right now, I won't be against that; if we have a healthy oversieve I can post-process this on my laptop which has only ~12 GB RAM available and some/most of my ~460M relations jobs won't quite fit there. I think stepping down one or two steps on VBITS would be a possibility here, but I will use a high memory machine in that case instead.

[swellman]

Originally Posted by kruoli

If we go by the numbers of Chris, we will need more relations, but I am willing to wait for our current WUs complete up to 10M outstanding, then I will download them and will give remdups4 a go to determine how much duplicates there are.
If you want to bump it right now, I won't be against that; if we have a healthy oversieve I can post-process this on my laptop which has only ~12 GB RAM available and some/most of my ~460M relations jobs won't quite fit there. I think stepping down one or two steps on VBITS would be a possibility here, but I will use a high memory machine in that case instead.

Let’s allow 145!-1_b to finish sieving before adding more Q - I don’t doubt more Q would be helpful but let’s see where we are first. 8m7_341 is sending out WUs now and should continue for another week or two so the NFS@Home machine is still running smoothly.

[kruoli]

145!-1 now has 331M unique relations (including pending ones by extrapolating). I would assume that this would be able to generate a matrix, but a big one (> 30M). Please correct me if this estimation is wrong.