Thoughts about upcoming LLRnet rallies

[Flatlander]

Originally Posted by Anonymous

Well, port 5000 is at ~490K (rounded up a little bit), and port 300 is at ~426K. That means ~110K remaining for port 5000 to reach the goal, and ~174K remaining for port 300. 110 divided by 174 is ~0.63, so that means there is 63% more work on 400<k<1001 (i.e. port 300) to be done than on 300<k<400 (i.e. port 5000).

174/110 = 1.5818
Times 6 because there are 6 times as many k, = 9.4908 times as much work.:surprised

9.4908/10.4908=0.9046
90.46% of work to do is on port 300.

Or am I being thick again?

[gd_barnes]

Originally Posted by Anonymous

Well, port 5000 is at ~490K (rounded up a little bit), and port 300 is at ~426K. That means ~110K remaining for port 5000 to reach the goal, and ~174K remaining for port 300. 110 divided by 174 is ~0.63, so that means there is 63% more work on 400<k<1001 (i.e. port 300) to be done than on 300<k<400 (i.e. port 5000).

This would not be 63% more work. It would be 63% more n-range. But the calculation is still incorrect on n-range. It should be 174 / 110 - 1 = 58% more n-range.

To prove the point, if you used your calculation where one drive was at 420K and the other drive 440K, you'd have 160 / 180 = 89% more work. Obviously that's not correct. There it would be 180 / 160 - 1 = 12.5% more work.

As for how much more work is remaining between the drives, you're forgetting the difference in # of k's. You could be exact and look at # of k/n pairs remaining but a good estimate would be to use the difference in the # of k's remaining to determine "work pieces" as follows:

Drive 1: (600-426) n-range * 6 times as many k's = 1044

Drive 3: (600-490) n-range = 110

1044 / 110 =
9.49 times as much work remaining or 849% more work on drive 1!!

Technically it would be slightly higher ratio because drive 1 was only sieved to P=5T vs. drive 3 that was sieved to P=6T, although that would only make an incremental difference. But then it would be slightly lower because the lower average n-range in drive 1 makes the average testing time faster...so 9.5 times more work should be in the ball park.

The case for doing a rally on port 5000 is only to quickly get that drive up to n=600K so that we can release all k's for n>600K and to avoid getting picked off by others that are not associated with the project searching some of the range. The case for doing a rally on port 300 is to make sure we hit n=600K by year end on that drive also.

Technically it doesn't matter which drive we do the rally on since the combined goal is to have both drives up to n=600K. The thing that might cause us not to reach the goal by doing the rally on port 5000 is if people then subsequently put a lot of resources on drive 3 for n>600K. I'm not too fearful of that with the current group we have doing searching for us although we'll find out for sure when I release 9 more k's for individual searching once we have the final testing range up to n=480K completed.

The main reasoning for pushing drive 3 is to avoid getting 'picked off' by others not associated with the project that don't report their primes as Primesearch. That range is more likely to be searched by others because lower k-values LLR faster because their fftlen does not increase until higher n-ranges. Also, as we release more k=300-400 for searching n>600K, my hope has been all along that we will bring in some people from RPS that like individual k searches but who otherwise wouldn't be interested in searches across a large # of k's.

For us, Benson is not a huge issue, only a minor annoyance. Although he 'picks off' our primes at times, he does report them as Primesearch so still helps the project. It just means that our regular contributors don't get to report them.

Although I would like to do a CRUS rally, with all the large contributors that we now have to this project and with 9.5 times as much work remaining on port 300, I think it makes sense to continue doing rallies on port 300.

If we do a CRUS rally, I think we should promote it in those project threads and but put a small sales pitch in for it here. We have several contributors there that do not do NPLB searches and vice-versa here. In other words, we do rallies for both projects, not one instead of the other. Probably rallies once every 1-2 months at CRUS are sufficient at this time.

I had made a statement in another thread that we should do rallies here only about once/month to allow Karsten/you to catch your breath in between rallies on administrative stuff. If we plan on doing them every 2-3 weeks like we did at the beginning of the project, then I'd suggest letting Karsten know so he can be prepared for frequent rallies. With the new contributors to the project that we got with the last rally, doing them every 2-3 weeks looks like a good idea now.


Gary

[gd_barnes]

Originally Posted by Flatlander

174/110 = 1.5818
Times 6 because there are 6 times as many k, = 9.4908 times as much work.:surprised

9.4908/10.4908=0.9046
90.46% of work to do is on port 300.

Or am I being thick again?

Dead on, we reached the same conclusion at almost the same time. lol

[mdettweiler]

Okay, you're right Gary and Chris--I was waaaaay off in my calculations.

So, it looks like doing port 300 again would be best--any objections to that?

Also, as for when we do the rally, I agree with Gary, we should probably do them every 2-3 weeks for now with all the new members joining. Karsten, is that OK with you? If so, what does everybody think of doing the next one on the weekend of May 17th?

We also had discussed previously about possibly doing a 48 hour rally sometime in the future--maybe we could do this one from 7 P.M. GMT on the 16th (Friday) to 7 P.M. GMT on the 18th (Sunday)?

[em99010pepe]

We should make a one week run...lol

[mdettweiler]

Originally Posted by em99010pepe

We should make a one week run...lol

Hmm, that's an idea. Maybe we should do one in the middle of the winter (i.e. winter 2008-2009), then all the extra heat will actually be helpful. (At this time of year I can only have my computer on about 90% of the time, since it's in my bedroom and otherwise it would make this too hot for me to sleep at night. )

[Flatlander]

Gary, I haven't seen your name on port 300 for a while but your % has gone up, are you using a different name?

[gd_barnes]

Originally Posted by Flatlander

Gary, I haven't seen your name on port 300 for a while but your % has gone up, are you using a different name?

I'm using lucky123 at the moment. Every once in a while, Adam will combine its stats with the gd_barnes stats on his web page. Since the rally, I've scaled back from 10 to 4 cores running port 300 to catch up on some CRUS and other personal efforts.

I don't want to say publicly why I'm using lucky123 but in a little while, it will no longer be necessary for me to do so and I'll change all of my machines back to gd_barnes. With the next rally, I'll use the gd_barnes ID again.


Gary

[gd_barnes]

Originally Posted by Anonymous

Hmm, that's an idea. Maybe we should do one in the middle of the winter (i.e. winter 2008-2009), then all the extra heat will actually be helpful. (At this time of year I can only have my computer on about 90% of the time, since it's in my bedroom and otherwise it would make this too hot for me to sleep at night. )

Very interesting! I like the idea of doing a week-long rally but I agree with Anon that we need to wait until winter to do one. The electrical cost is bad enough with no A.C. running. Do we have any folks from the Southern Hemisphere here? Obviously winter is coming up for them.

This is painful to state but...I just looked at my electrical usage per day since I got 5 of my 6 quads running. It jumped from $1.50/day to $4.50/day all with no A.C. or heat running and with the machines sitting in a nice cool basement (that for some reason is not quite as cool now, lol). OUCH!! And electricity is generally relatively inexpensive in the midwest.

So that's ~$90 / month to run all cores on 5 quads 24x7. I guess it'll be ~$108/month once I get the 6th quad finished. I just got the new motherboard for it and will get it running by Thursday.

Beyond, Ironbits, Linnert, and any other folks with numerous personal machines running: Is $18/month per quad typical electrical usage?

I had expected about 2/3rds of that cost, i.e. ~$10-12/quad.

Ugh!


Gary

[gd_barnes]

Originally Posted by Anonymous

Okay, you're right Gary and Chris--I was waaaaay off in my calculations.

So, it looks like doing port 300 again would be best--any objections to that?

Also, as for when we do the rally, I agree with Gary, we should probably do them every 2-3 weeks for now with all the new members joining. Karsten, is that OK with you? If so, what does everybody think of doing the next one on the weekend of May 17th?

We also had discussed previously about possibly doing a 48 hour rally sometime in the future--maybe we could do this one from 7 P.M. GMT on the 16th (Friday) to 7 P.M. GMT on the 18th (Sunday)?

With Flatlander not having internet for 2 weeks and with May 24th-25th being the Memorial Day weekend, that might be a better weekend to run a 48-hour rally. Anon and Karsten, will you be around that weekend?

Will other people be around over the Memorial Day weekend? Pardon my ignorance of holidays in other countries but I don't know if most of Europe celebrates the U.S. Memorial Day, which is the last Monday in May, where most of us are off work (not that others being off work matters for me, lol). A calendar that I have shows "Memorial Day (U.S.)" so I'm thinking that may be specific to the U.S.


Gary

[Flatlander]

I will be taking the kids' Athlon PC on holiday to the coast for the fortnight. As we will be overlooking the sea I will be hooking up my new camera to it to do some time-lapse photography. I also need the PC to download astronomy photos from my new telescope.
I will leave the Quad here running at stock speed. (With internet access.) Like I said above, if I have two LLRNET clients running on each core then if one stalls the other will, hopefully, get 100% of that core.

In other words my Quad should be on port 300 until at least May 23rd.

Can someone remind me how to set the WUCacheSize > 100 please. (I read it somewhere.)