Tracking GIMPS progress with scripts

[jasong]

Lately, I've become fascinated with tracking GIMPS progress. Mostly, I've been watching (1) the leading edge(basically, I'll go to the end of the file and take a gander at the last number listed), (2) trailing edge of double-check, which is basically the very beginning of the file.

I'd like to add a third one, tracking the trailing edge of first pass, but the Edit feature of my browser doesn't seem to be up to the the task. Any ideas on how I could accomplish this?

[petrw1]

Does this sort of help?

I simply counted "Cleared" records by Exponent Millions by month since Sept 2006.

I also have individual sub-reports by LL, DC and Factoring.

I also have corresponding graphs.

From the attached chart you can see 4 "lumps":
1. One moving from 17M to 19M: Current Double Check range
2. One decreasing in volume at the 32M range: First Time tests below 10M digits
3. One moving from 35M to 38M: Current first time LL test above 10M digits.
4. One moving from 39M to 42M: Current factoring range.

You can also see at the top the total per month is actually decreasing. This surprises me somewhat. While granted larger exponents take longer to LL test, the hardware should be increasing in speed at a similar rate, and hopefully the number of members grows or at least stays stable.

[davieddy]

Originally Posted by petrw1

You can also see at the top the total per month is actually decreasing. This surprises me somewhat. While granted larger exponents take longer to LL test, the hardware should be increasing in speed at a similar rate, and hopefully the number of members grows or at least stays stable.

I have noted a slight decline in participation during
this period.
But the decline in tests returned is largely
accounted for by at least one increment
in FFT size, and larger exponents so the tests are now taking
significantly longer than in Sep 2006. (Your table shows this
clearly).
Note especially the jump from first time tests with exponents <33M
to those >36M, due to the lowest 10M digit numbers having been
tested long ago.

David

PS As for hardware improvements, I get the
impression that the exploitation of multicore
processors is not without its problems.

[petrw1]

We are very close to another big jump in FFT and testing time at 40,250,000. Not sure how much longer the tests will take but I can guess 33% since the points awarded will jump from 8.6 to 11.5.

Granted multi-core hasn't caught on yet but I would assume on any given day dozens of GIMPS PCs are being replaced or upgraded ... maybe from a P-3 866Mhz to a P-4 2.8Ghz (as I did last month) or from a P-4 1.7Mhz to a P4 2.4Ghz as I did a couple years ago OR adding friends and family computers; I added 3 in the last few months.

[davieddy]

Originally Posted by petrw1

We are very close to another big jump in FFT and testing time at 40,250,000. Not sure how much longer the tests will take but I can guess 33% since the points awarded will jump from 8.6 to 11.5.

Yes. According to GIMPS status page the iteration time
increases by 33%. Note that the FFT boundaries are
slightly different for the fastest computers.
I think the rate of processing has remained around 24 Teraflops/s
during the last few months.

[jinydu]

Originally Posted by davieddy

PS As for hardware improvements, I get the
impression that the exploitation of multicore
processors is not without its problems.

What do you mean by that?

[TimSorbet]

Originally Posted by jinydu

What do you mean by that?

While I'm not him...I'd guess that he meant that with multicore CPUs, people will be using the less efficient method of having multiple cores work on one exponent so that they get each one done faster.

[davieddy]

Originally Posted by jinydu

What do you mean by that?

Lots of posts to that effect.

[jasong]

Originally Posted by Mini-Geek

While I'm not him...I'd guess that he meant that with multicore CPUs, people will be using the less efficient method of having multiple cores work on one exponent so that they get each one done faster.

I was thinking more along the of the fact that bandwidth is becoming a real problem. I'm thinking that a lot of people involved in GIMPS will either crunch more than one project simultaneously on the hardware, because of latency problems(and factoring can only take so much new blood before you're insanely far ahead of first-pass). I think what people will find the most rewarding, IMAO, is taking the Intel quad and running one first-pass exponent on cores 0 and 1, and another first-pass exponent on 2 and 3.

JMO.

[davieddy]

Originally Posted by petrw1

Granted multi-core hasn't caught on yet but I would assume on any given day dozens of GIMPS PCs are being replaced or upgraded ... maybe from a P-3 866Mhz to a P-4 2.8Ghz (as I did last month) or from a P-4 1.7Mhz to a P4 2.4Ghz as I did a couple years ago OR adding friends and family computers; I added 3 in the last few months.

Talking of upgrades, my second hand PII 350MHz expired
shortly after it had verified that M17,105,059 was composite
in 5 months

I have now acquired a second hand PIII 800 MHz but I am
reluctant to thrash it with another LLtest

David

[petrw1]

Originally Posted by davieddy

Talking of upgrades, my second hand PII 350MHz expired
shortly after it had verified that M17,105,059 was composite
in 5 months

I have now acquired a second hand PIII 800 MHz but I am
reluctant to thrash it with another LLtest

David

My first GIMPS PC was a PII 400Mhz ... it ran for 15 months on M33xxxxxx and last reported in at 99.8% complete before my son ran something that swallowed up every bit of memory in an instant. The PC crashed so quickly that when I powered down and up again Prime95 determined it had to start from the beginning. I cried then I cancelled that test because in the meantime I had purchased a P4 1.7Ghz that could do the same test in 1 month. I relegated the P400 to Double checks; in the 11M range.