[SGVLUG] Samba performance issue

[Rae Yip]

69 Celcius seems a bit high. On the other hand, if you're getting
decent performance except for every 20 minutes, I suppose that's not
the problem.

I was hoping that if your average throughput was high enough, then
increasing the buffer in VNC would decouple your sensitivity to
throughput jitter.

But I suppose that only solves playback and not the editing slowness.
It's not clear to me whether this is a network problem or a drive I/O
problem. Until you do comparative testing between local performance
vs. over the network, it's hard to figure out what needs fixing, so
you should start there.

You might also try seeing if you run into the same issue with NFS as
with Samba, if it turns out you're not seeing local performance dips.

-Rae.

P.S. Journaling improves data integrity but will impact performance
depending on the type of accesses.



On Thu, May 22, 2008 at 12:14 AM, Matt Campbell <dvdmatt at gmail.com> wrote:
> Oh Yuck!
>
>  I formatted the arrays ext3, it looks like it's really slow compared to
>  ext2.  Is that generally correct?  Is journaling the main difference between
>  versions 2 and 3?
>
>  I installed and ran iozone.  Neat program, but it will take a bit to get my
>  head around which things are interesting to test for.  Attached please find
>  an XML spreadsheet of the first data run I did.
>
>  There is something going on, while I am streaming data from the sever the
>  network usage looks like a sine wave....
>
>
>  Matt
>
>  > -----Original Message-----
>  > From: sgvlug-bounces at sgvlug.net [mailto:sgvlug-bounces at sgvlug.net] On
>  > Behalf Of Dan Kegel
>
> > Sent: Wednesday, May 21, 2008 9:26 PM
>  > To: SGVLUG Discussion List.
>  > Subject: Re: [SGVLUG] Samba performance issue
>  >
>
>
> > On Wed, May 21, 2008 at 8:16 PM, Matt Campbell <dvdmatt at gmail.com>
>  > wrote:
>  > > I can write a simple program on a Linux or Windows client machine to
>  > make
>  > > simple random file requests and time the responses, but would this
>  > really be
>  > > indicative of streaming performance?  I could open a large media file
>  > > (larger than RAM on server and client combined) and request block
>  > after
>  > > block and graph response times and dropped blocks.
>  > >
>  > > What type of testing do you think should be done to deliver a useful
>  > > benchmark?
>  >
>  > http://iozone.org/ has a read latency measurement, maybe that's a
>  > start?
>