Eric Rosello <rosello@gfsimdoh.com> wrote:> It looks like
> this performance problem is also visible if you try to access any file
> already open by another user (at least in my case). If anybody wants to
> try, just open a large file (1Mb or 2 Mb) on one machine, and then from
> another machine, copy this same file to another location (so you access
> it in read only). In my case there's a big difference in speed. If
> nobody is accessing the file, then the speed is more or less what you
> will expect for a 10Mbps network. But if somebody has the file already
> open, then it takes much longer, up to 8/10 time longer.
I think you're seeing the initial access times, which are worse
than subsequent access times...
If you open a file for update, and if you're the only one,
you will copy it once from the server to a cache on
your machine. If you're not the first one, you'll open the file,
try to read, and be suspended while the first user flushes
his cache back to the server. Then you'll run at network speeds.
If you're trying to model this with file copies, you'll
see the initial slowdown, but never find out how fast or slow
it is on all other reads!
You may want to write a trivial programs which opens, seeks
and reads several times, reporting the open time and
each subsequent read-time. Posting those results should
help the Samba team decide what can be done to help.
--dave
--
David Collier-Brown, | Always do right. This will gratify some people
185 Ellerslie Ave., | and astonish the rest. -- Mark Twain
Willowdale, Ontario | http://java.science.yorku.ca/~davecb
Work: (905) 415-2849 Home: (416) 223-8968 Email: davecb@canada.sun.com