Smarandache prime(s)

[rogue]

Are you running your factors thru pfgw to verify that your sieving program is working correctly?

[Batalov]

Indeed I am, for an independent validation.
Here are some quite big examples:

$ ../../M/pfgw377/pfgw64.exe  -o -d -q"Sm(937357) % 260889573299"
PFGW Version 3.7.7.64BIT.20130722.Win_Dev [GWNUM 27.11]
Sm(937357) % 260889573299 is Zero (0)

$ ../../M/pfgw377/pfgw64.exe  -o -d -q"Sm(950083) % 350500327669"
...is Zero (0)
$ ../../M/pfgw377/pfgw64.exe  -o -d -q"Sm(789529) % 350758089373"
...is Zero (0)

[Batalov]

In case anyone is curious, I've set quorum to 1. Everyone will get a fresh first time test (good luck to everyone!).
DC will commence later...

[pinhodecarlos]

Hi Serge,

I am averaging 36500 sec per candidate on my i7 4800MQ machine.
Carlos

[Batalov]

Yes, this sounds pretty good. This will go up, though. The run-time is roughly O(n^3) for the range that I've already run.

The i3-5010U has a tiny cache, so it takes a performance hit. I am pretty happy with the cheap laptop that I grabbed from Amazon recently ($349 + $22 for the second stick of memory).
The EC2 Haswells (c4.8xlarge) are not terrifically fast but they are relatively cheap to play with: < 2c / hr-core.

[Batalov]

Plenty of prime candidates are still available. Prime cuts, expertly sieved to 0.36T (and still sieiving).
Don't miss your chance!

[LaurV]

Took one. Will take more.
Can we multithread? (I mean, all cores working for the same assignment, and not running more workers, for that I am ok with multiple instances)

Edit: 3M9 bits? Well, if I find a prime, it will be not only the biggest prime I ever found, but also one of the biggest primes I ever dreamed to find... I never thought too seriously to find million-digits primes...

[LaurV]

Sorry for new post, time limit.

Took another.

How can I do to have the Prime95's equivalent of "NoMoreWork=1"?
(i.e. when the current assignment is finish, it will exit and/or wait for me, without reserving new work).

I looked into the ini file but I can't find the option, the things seems related to "ctrl+c" only, or they "shut down" things which I don't know if it means the program or the computer. I will not want my computer shut down when the assignment is finished. Just not more work being reserved. I will only reserve work if I know I can do it, and not be in the situation to return it back or worse, forget about it at all. This is not something I do every day.

edit: your "Found a factor!" caption of the picture should better be "Find a prime!"

[rogue]

This is what you need to set in prpclient.ini:

// stopasapoption= tells the client that it needs to be shutdown automatically, i.e. without
// a CTRL-C. It is evaluated after each test is completed. It should be 0 upon startup.
// The accepted values are:
// 0 - Continue processing work units
// 2 - Return completed work units and abandon the rest
// 3 - Return completed work units (keep the rest)
// 6 - Complete all work units and return them
stopasapoption=0

Use option 2. You can edit while the program is running.

No, you cannot multithread.

[LaurV]

Thanks. I don't want to abandon anything, just finish all work I have and don't request new work. Option 6 is closer to what I want. Meantime I found out by experimenting that I can also use ctrl+c then input the option to finish the work and shut down, which is just a bad phrasing for "stop the program" (and not the computer).

P.S. Took another 4, about 3 hours ago. They will be finished in about 20-25 hours.

[Batalov]

I pre-filled all start/stop options to 9/0 on the deployment tarball to the EC2 (spot) nodes.

machineid=EC2-Or2
instanceid=3
server=SmarPRP:100:1:99.121.249.54:1200
pfgwexe=../pfgw64.exe
...
normalpriority=1
// startoption= tells the client what to do when starting up and there is work that has
startoption=9
// stopoption= tells the client what to do when it is stopped with CTRL-C and there is
//    9 - Do nothing and shut down (presumes you will restart with startoption=9)
stopoption=0
// stopasapoption= tells the client that it needs to be shutdown automatically, i.e. without
stopasapoption=0

Because they frequently get cancelled, the state is preserved on an attached drive. So if set up with option=9, I can create a new spot node, attach the volume, login, mount, cd, and restart all threads. Binding each prpclient to a separate core (# 0 - 17 on c4.8xlarge) is also helpful on EC2 or else OS mixes and matches them on 36 logical cores and some run at half speed (running 36 has no advantage - that's tested). Of course there are other ways, less mechanical, and they have some new spot manager launched recently.