ORIGINAL POSTING (extract)
> We have one user with a persistent performance problem.
>
> He has a program (written in Fortran) which does very many
> small writes to disk. It is disk-bound most dramatically, with
> ps -v showing 0% CPU, and the disk performance meter
> remaining steady at about 50% for the whole run time.
>
> I have suggested tmpfs and the buffer option in Fortran.
> We find these do not cure it separately or together.
>
SUMMARY
The user's program used a large direct-access file for both reading and writing.
It WAS arranged so that he read it contiguously and wrote to it sparsely.
Now he does it the other way round.
He thought of it himself - *now* after weeks of bothering me.
It makes all the difference !
Thanks to Jon Green (jonsg@diss.hyphen.com) for a suggestion concerning mmap,
not used in the event.
--________ ___________________________________ /\ | ______| /\ | | / \ | | / \ | Peter Allan | / /\ \ | |______ / /\ \ | Email : peter.allan@aea.orgn.uk | / /__\ \ | ______| / /__\ \ | Phone : (44) 925 252684 | / ______ \ | | / ______ \ | Fax : (44) 925 252390 | / / \ \ | |______/ / \ \ |___________________________________|/_/ \_\|_________/ \_\ T E C H N O L O G Y
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:09:02 CDT