> Marty Leisner wrote:
> >
> > smbfs seems to took a packet/file to do an ls in 2.1.61
> > (its doing a getatr for every file).
> >
> > In 2.0.31, it does a TRANS2_FINDFIRST, and gets all this information
> > in a single smb.
>
> Present smbfs uses the dentry layer for caching inode information,
> rather than caching it again in smbfs. If you do an ls without
> requesting file info, the reply contains just file names, so it takes
> fewer message packets. Operations that use getdents (e.g. make) will
> also run faster, as we're not requesting info that can't be used.
>
> The directory code still uses TRANS2_FINDFIRST/FINDNEXT, it just
> requests only the minimal info required by a readdir call.
>
> If you have specific preformance benchmarks where you think the current
> smbfs isn't working fast enough, I'd be happy to look into it.
Bill,
I have been casually using smbfs in 2.1.61 to mount shares on an Alpha UDB
running WinNT Workstation 4.0. Writes are dog-slow (~250-500 KB/sec), but
reads are acceptable (~500-700 KB/sec.) given that it's 10Base ethernet.
The real annoyance is that I cannot unmount the share from the Linux box!
It just complains that the file system is busy, which I've taken pains to
ensure is not the case.
Anything you'd like me to try?
Steve