Discussion related to cado-nfs

[paul zimmermann <Paul.Zimmermann@inria.fr>]

       Dear Valentin,

> From: Valentin Janser <valentin@wans-janser.de>
> Date: Wed, 15 Apr 2020 20:23:38 +0200
> 
> Hello Paul,
> 
> 
> I now have two machines which are similar to each other.
> 
> The only difference is that one of them has 1 E5-2680, the other one has
> 2 E5-2680.
> 
> I start the computation with the command ./cado-nfs.py <number>
> server.ssl=no for both machines.
> 
> 
> The machine with 1 CPU says:     Set tasks.threads=16 based on detected
> logical cpus
>                                                                 
> tasks.polyselect.threads = 2
>                                                                 
> tasks.sieve.las.threads = 2
> 
> When sieving, I can see 8 processes named "las" using nearly 100 % of 2
> logical threads, that makes a near 100 % usage of all 8 cores/16 threads.
> 
> 
> The machine with 2 CPUs say:    Set tasks.linalg.bwc.threads=16 based on
> detected physical cores
>                                                          Set
> tasks.threads=32 based on detected logical cpus
>                                                                
> tasks.threads = 32 [via tasks.threads]
>                                                                
> tasks.polyselect.threads = 2 [via tasks.polyselect.threads]
>                                                                
> tasks.sieve.las.threads = 2 [via tasks.sieve.las.threads]
>                                                                
> tasks.linalg.bwc.threads = 16 [via tasks.linalg.bwc.threads]
>                                                                
> tasks.sqrt.threads = 8 [via tasks.sqrt.threads]
> 
> There are 16 processes named "las" also using nearly 100 % of 2 logical
> threads (according to the command "top") which would result in near 100
> % usage of all 16 cores/32 threads.
> 
> 
> The machine with 1 CPU  has a CPU time of 10574 s, a Wall Clock Time of
> 911 s.
> 
> The machine with 2 CPUs has a CPU time of 24037 s, a Wall Clock Time of
> 984 s.
> 
> 
> Why is the machine with half the theoretical performance faster than the
> other server?
> 
> For me, it seems like the second CPU is not calculating anything...
> 
> Am I doing something wrong?

this is indeed very strange. Which version of cado-nfs are you using?
Can you post the full console log of both runs?

> Also, you have a table on your website. According to that, you factor a
> 120-digit-RSA-number (probably the one from wikipedia) in roughly 2
> hours on a dual socket E5-2650.
> 
> I think that your calculation also doesn't use all cores to their full
> potential; with my single-cpu E5-2680-server I can factor a
> 120-digit-RSA-number in roughly 2 hours aswell.

the timings you refer on the web page are for cado-nfs 2.3.0.
Did you use that version?

Best regards,
Paul