M929 progress: Sure, drinking kills brain cells - but only the slow ones!

[J.F.]

Originally Posted by fivemack

Should start sieving in a couple of days - getting statistically reasonable estimates for comparing strategies does appear to require about two CPU-weeks per parameter set at this level, and I don't have that many CPUs.

I would really like to get back into this business. Years ago, I had much fun abusing University spare cycles to do some considerable sieving with you guys. I miss those times, I miss my cluster access. So I am thinking of making my own tiny cluster, nicely virtualized and such, running say 4 cores on GIMPS, 4 cores SOB, and whatever else on custom jobs such as you are about to start. Maybe even do some LA eventually :)?!

I still need to figure out how to spend my money: what kind of system and virtualization software etc. Bit of a hardware noob... Suggestions are welcome! [If plenty, split to separate thread?]

A friend of mine suggested virtualization and the ease of plugging in extra hardware whenever money needs to be thrown away (...). I have some to spare now.

[henryzz]

There were no large enough primes in the dependencies. Running with 50M relations only made things worse. All cycle lengths were 1. Sieving didn't take that long so I am going to start off a larger job. The largest factorization the siever did on this one was 50 bits. I don't have the disk space on my linux partition to keep the current dataset as well.

[henryzz]

Successful test with poly:

n: 158657932830335643672510832547714307838661647397769627056857550567543510512568504872884888489508116023549485662557130756632182845328127
m: 3549077213064806756794057860328127
deg: 4
skew: 1.19
type: snfs
c4: 1
c2: -4
c0: 2
rlim: 1000000
alim: 1000000
lpbr: 35
lpba: 35
mfbr: 80
mfba: 80
rlambda: 2.9
alambda: 2.9

I sieved 600K-1M

[fivemack]

Originally Posted by J.F.

I would really like to get back into this business. Years ago, I had much fun abusing University spare cycles to do some considerable sieving with you guys. I miss those times, I miss my cluster access. So I am thinking of making my own tiny cluster, nicely virtualized and such, running say 4 cores on GIMPS, 4 cores SOB, and whatever else on custom jobs such as you are about to start. Maybe even do some LA eventually :)?!

I think virtualisation is a red herring if you're talking about clusters; we're running single applications which don't have a stack of dependencies the size of Belgium, just run them on the bare OS.

I had a cluster and replaced it with a single ludicrous machine; this let me do much bigger problems, because the interconnect in a quad-core Opteron is much better than any affordable cluster fast-interconnect, though the supply of bigger problems is small enough that it spends a lot of time being embarrassingly parallel.

I might supplement the ludicrous machine with a Haswell or two when they come out in the summer, but mini-tower PC cases take up a lot of space and I haven't found a decent cheap small case ... you don't need drive bays when you're using NFS and your OS is on a USB stick.

[ewmayer]

Originally Posted by fivemack

I might supplement the ludicrous machine with a Haswell or two when they come out in the summer, but mini-tower PC cases take up a lot of space and I haven't found a decent cheap small case ... you don't need drive bays when you're using NFS and your OS is on a USB stick.

Hi, Tom - this may be of interest to you:

How to Mount Two ATX Motherboards in One Case | eHow.com

I'm mulling what to do when it comes time to order a Haswell-based quad MoBo in a few months - default option is to swap out my current Sandy Bridge MoBo for the new one, but as the PS is beefy enough [650W] to run 2 quad-cores, it would be very neat to be able to pack them into the same ATX case.