> We've had up to about 17MBytes/sec on a Synergy VGM5 (the driver had
> been modified by Synergy), and about the same on a MVME2307 with SYM875
> (we had to get hold of the source code and modify it to get that
> performance). I've never used the 895, but if it is using the same
> driver as the wr$ driver for the mv2307 there's a lot of tweaking
> possible to get big performance improvements (both increasing data
> transfer rate and reducing pci usage).
<nod>. Just now I timed scsi performance under both Win2k and vxWorks on
the same hardware, and Win2k performed it's 4MB write in about 100ms,
which is pretty decent. VxWorks performed the same write in 400ms. I dove
in with windview to try to see what's going on, and found that the SCSI
driver's interrupt handler gets called a few times at each of THREE points
during a write-operation. These points are at t=0ms, t=100ms and t=400ms.
The t=100ms point is interesting, as that's EXACTLY how long windows takes
to write the data. I'm starting to suspect that the scsi driver is
actually writing the data at a reasonable speed, but then is doing
SOMETHING else before it returns successfully. *shrug*
That's a bummer that we might need to get the code for the driver and
tweak it. Oh, well.
Thanks for the reality-check,