Alex Granovsky
gran@classic.chem.msu.su
> May be, you have still any advice as a reserve :-)?
yes I have... you can try
$system decomm=.f. wsctl=.f. $end
vs.
$system decomm=.t. wsctl=.t. $end
on some large job to check if this affects performance.
Regards,
Alex
On Fri Apr 2 '10 1:58pm, Vyacheslav wrote
-----------------------------------------
>Hi,
>with mxgsum=8000 performance strongly decreases (approximately in 2 times for 24 nodes). It was the last job in first series. Probably I should not exceed this value (in decimal) during tuning?
>Further I'll do runs as you have suggested, Alex, and then I'll try to vary other tuning options, mentioned above in this discussion.
>Thanks for brilliant recommendations, dear Alex! May be, you have still any advice as a reserve :-)?
>Thanks to all for various help!
>---------------------------------------------------
>
>
>On Fri Apr 2 '10 1:24pm, Alex Granovsky wrote
>---------------------------------------------
>>Hi,
>>glad to know that!
>>00400 means simply 400 i.e. leading zeros do not affect the value.
>>I'd suggest you to check the sequence I have suggested - i.e.,
>>the consequent powers of two: 512, 1024, 2048, etc...
>>Regards,
>>Alex
>>
>>
>>On Fri Apr 2 '10 12:49pm, Vyacheslav wrote
>>------------------------------------------
>