Sieving drive for k=1003-2000 n=500K-1M

[gd_barnes]

Originally Posted by IronBits

If I were to take P=40-50T range...
If you break it up to go across 4 x 8 cores = 32 * 3.5 GHz cores
How long would that take?

I'm not sure yet that we will need to go all the way to P=50T. What I meant was that we may need to sieve up to BETWEEN P=40T and 50T somewhere.

But...if you would like to take the range of P=30T-40T, that would be excellent! We'll almost definitely need that.

Let's see: P=250G (0.25T) on one 2.6 Ghz core takes right around 15 days right now so on 32 3.5 Ghz cores, you could do:

32 * 0.25T * 3.5 / 2.6 = 10.77T in 15 days.

Well, isn't that convienient that it comes right out close to a P=10T range? Therefore assuming that you have a 64-bit machine, O.S., and the 64-bit version of sr2sieve (which I'm sure you do, lol), I'm confident that you could complete P=30-40T in 15 days on 32 cores.

If you'd like that range, I'll reserve it for you and then send you the new sieve file late tomorrow afternoon after we have all of the factors up to P=20T removed (at the same time that I send you another file for port 8000). You could feel free to take a break for the rally.

If Bruce does P=20T-30T, you do P=30T-40T, and the optimum sieve depth turned out to be P=50T, I'm sure that between you guys, me and the rest of the NPLB group, we could all get it done by early June, which is what I'm hoping for. (I think the 10th drive will be nearing n=600K at that point.)


Gary

[IronBits]

After the rally, sign me up for 30-40T then.

If you or Max could divide it up for 4 computers with 8 cores each, then I'll fire em up.
(like last time) except this time, I'll let it finish correctly

[mdettweiler]

Originally Posted by IronBits

After the rally, sign me up for 30-40T then.

If you or Max could divide it up for 4 computers with 8 cores each, then I'll fire em up.
(like last time) except this time, I'll let it finish correctly

Okay, I'll see about setting you up with a "care package" like last time as soon as the p=20T sieve file becomes available. Remind me when the time comes if I forget...

[gd_barnes]

Originally Posted by IronBits

After the rally, sign me up for 30-40T then.

If you or Max could divide it up for 4 computers with 8 cores each, then I'll fire em up.
(like last time) except this time, I'll let it finish correctly

Excellent! Thanks for helping out.

Bruce, although I put you down for P=20T-30T, after you get your machines rolling with the next P>20T range (I'm assuming 250e9 per core), let me know what specific range you are running since I'm not clear exactly how many cores you will be sieving with. I will likely put the end of your range up to P=30T up for public reservation. Therefore, ultimately one of the ranges running on your machines at one time may end up having a big gap in the middle, perhaps like P=28T-30T and 40T-42T.

Max, I'll send the P=20T sieve file to you and you can include it in your care package to David. As for dividing P=10T by 32 cores, it would be 0.3125T or 312.5G or 312500M per core. Since that's a bit messy, you might make half of them 312G and the other half 313G. If you do that, I would suggest making the higher ranges 313G since they sieve a tad faster. So the 1st core would be 30000e9-30312e9, 2nd core, 30312e9-30624e9, etc. That should put them done all at very close to the same time.


Gary

[mdettweiler]

Originally Posted by gd_barnes

Max, I'll send the P=20T sieve file to you and you can include it in your care package to David. As for dividing P=10T by 32 cores, it would be 0.3125T or 312.5G or 312500M per core. Since that's a bit messy, you might make half of them 312G and the other half 313G. If you do that, I would suggest making the higher ranges 313G since they sieve a tad faster. So the 1st core would be 30000e9-30312e9, 2nd core, 30312e9-30624e9, etc. That should put them done all at very close to the same time.

Okay, that sound good. Having the ranges slightly differing in size as you described should definitely make things a little easier to keep track of, and any resulting differences in runtime should surely be within a normal range of variation that occurs due to other processes on the system and other such factors.

[Brucifer]

Mr. Barnes, you will find in your email inbox 19850e9-20000e9 in gzip format.

I have now loaded in 20000e9 to 25000e9 in 250G ranges, covering 20 cores. These should complete on Apr 23/24.

So let me know what your plan is for the remaining ranges when you decide please.

[gd_barnes]

P=19750G-19850G is complete.

We are now complete up to P=20T. A new sieve file has been sent to everyone.

Bruce, since it appears that you are running 20 cores for sieving, would you want to just put a P=500G range in each of them? That would take care of the P=20T-30T range and would make things basically "maintenance free" on your end. That would take you just a little under 30 days. I've noted the range as reserved by you with an ETA of May 15th, allowing 5 days for unforseen problems and running the rally.

I'm now preparing the next file for port 8000 and will run some tests to verify the optimum sieve depth.


Gary

[gd_barnes]

Well, bad news followed by good news...

Bad news:

The optimum sieve depth for the entire range is definitely right at P=60T. (ouch) This is based on a current factor remove rate of 310 secs. and a test at 70% of the n-range (n=850K) taking 920 secs.

Good news:

We're not testing k=1003-1400 at the current time and the continued sieving of the n=500K-600K range is only for future double-checking.

Although we will NOT be removing the above portions of the sieve file, because the double-checking of both of those ranges will be well into the future when computer capacity is much greater, I feel that it should be taken into account.

Therefore we have:

3/5ths of the k-range * 4/5ths of the n-range = 12/25ths of the entire file will be tested in the foreseeable future.

The square root of 12/25 = ~.7

60T * .7 = 42T

So let's just round it up and sieve to P=45T. If we double-check in the future, if we want, we can sieve it further because capacity and speed will be much greater.

Sieving with today's resources to P=60T is overkill. Even if we were testing the entire range, it will be so long before it is complete that we could always sieve further in the future for the higher n-ranges.


Gary

[IronBits]

Sign me up for 30-45T then.
What's an extra day or two?

[PCZ]

David
Thats a lot of sieving.
I offer to take a more modest slice, 45-50T.

[IronBits]

Thanks for the tip Brian.
I'm just trying to help out a little.
If that's too big a bite, then Gary and Max can reduce my numbers and slice it up any way it best helps them out. (I'm clueless)

Getting excited about the 2 day rally!