Serving the Quantitative Finance Community

 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Parallel RNG and distributed MC

July 27th, 2012, 2:00 pm

Thanks. In the link I posted, does that mean somebody did what you are suggesting (scaled up to 17024 cores)by creating ~ 100 of these hight cpu instances with "extra multi-machine communication"?
Last edited by Alan on July 26th, 2012, 10:00 pm, edited 1 time in total.
 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Parallel RNG and distributed MC

July 27th, 2012, 2:13 pm

QuoteAmazon has groups 20 machines glued together with special hardware and software that makes it behave like a single big machine with 160 cores. There is no need to aggregate in a special way. Wow -- I didn't realize that. Is that discussed in the docs? Is that a proprietary amazon invention or something that hasbeen developed by the open source community and is available to anybody?
 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Parallel RNG and distributed MC

July 27th, 2012, 7:21 pm

I see -- thanks for the hint. A little googling shows that amazon EC2 is running Xen, something new to me. (This is really a nice project -- I am learning a lot! -- thanks to outrun and others)
Last edited by Alan on July 26th, 2012, 10:00 pm, edited 1 time in total.
 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Parallel RNG and distributed MC

July 28th, 2012, 1:27 pm

Thanks. It compiled and ran with the intended final result.However, the intermediate prints are garbled together.(I am running it as a console app)----------------------------------------------------------------------------------------------------------------------------- job 1 finished by thread job job 123 result = finished by thread finished by thread 0.24983702 result = result = 0.2436830.242275job4 finished by thread 3 result = 0.243981job 5 finished by thread 1 result = 0.246757job job 67 finished by thread finished by thread 20 result = result = 0.2453820.242287job 8 finished by thread 3 result = 0.241372job 9 finished by thread job job 110 result = 110.249875 finished by thread finished by thread30 result = result = job 0.2368820.24208512 finished by thread 2 result = 0.245032job 13 finished by thread job 014 result = finished by thread job 0.246988115 result = finished by thread 0.2397813 result =0.24651job 16 finished by thread 2 result = 0.244845job 17 finished by thread job 018 result = finished by thread job 0.246563119 result = finished by thread 0.2453483 result = 0.241295job 20 finished by thread 2 result = 0.247558final aggregate estimate is 0.244417 (0.00328516)Press any key to continue . . .
 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Parallel RNG and distributed MC

August 4th, 2012, 12:35 pm

Thanks -- it looks very nice. I am travelling now, but should have a chance to try a compile tues or weds.
 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Parallel RNG and distributed MC

August 8th, 2012, 2:45 pm

Got to it. Apart from having to manually unzip an boost file, all compiled and ran well.For the -S option, 0.1 is apparently too small for my machine and the reported time was quite a bit higher (0.28 secs). But trying -S n for n=1,2,..,5 worked ok with the reported time just slightly over n secs.Clearly example 1 will have machine-dependent results.I am not sure if example 2 results are supposed to be machine independent or not.Here are my results for that one: parallel_monte_carlo -T 8 -J 16 -Vjob 1 finished by thread 1 result = 0.246723job 2 finished by thread 3 result = 0.239763job 3 finished by thread 2 result = 0.23102job 4 finished by thread 0 result = 0.242088job 5 finished by thread 7 result = 0.249677job 6 finished by thread 4 result = 0.243666job 7 finished by thread 5 result = 0.264276job 8 finished by thread 6 result = 0.229672job 9 finished by thread 1 result = 0.248875job 10 finished by thread 0 result = 0.227201job 11 finished by thread 3 result = 0.248969job 12 finished by thread 2 result = 0.249528job 13 finished by thread 7 result = 0.233896job 14 finished by thread 5 result = 0.249619job 15 finished by thread 6 result = 0.237287job 16 finished by thread 4 result = 0.240146result=0.24265 err=0.00239089 threads=8 jobs=16 samples=32000 seconds=2.28125
Last edited by Alan on August 7th, 2012, 10:00 pm, edited 1 time in total.