Hi Mike!

[..]
 PH>> HOPFS can't reach up to HPFS performance by design. 
 MB> Optical media of the kind that HPOFS uses is organized in
 MB> cylinder-head-sector arrangement on the physical media, and most such
 MB> drives are accessed via SCSI using RBA counts.  Aside from the device
 MB> type code returned in response to a SCSI Inquiry, these drives look
 MB> just like hard drives even from the SCSI point of view, with the
 MB> exception that you can access only one side of the platter at a time.
You're right. But HPOFS uses different caching mechanisms optimized for
_optical medias_. Whereas HPFS uses optimized caching algorithms for hard
drives - and in particular my Syquest drive looks much more like a hard
drive than a optical drive. So HPFS's performance is unreached so far.

[FS_UNMOUNT, FS_FLUSHCACHES]
 MB> You could write your own IFS to do this.  There is no reason in
 MB> principle why it could not be done, and I would be happy to do it for
 MB> you if you wrote a sufficiently large check.
I've got certain doubts on that:the CDI-caches which are maintained between
kernel and IFS. I ran into the problem that CDI caches are not invalidated
100% reliably under unfortunate circumstances. I was unable to track this
down completely.
Second, to enhance OS/2's IO capability regarding the IFS, I rather dislike
to implement new features via FSCtl. I _really_ think IBM should enhance
the interface by a few standard functions. Anyway, I fear I can't handle
writing my own IFS ... :)

[..]
 MB> You would need a sideways control utility to send IOCtl directives
 MB> down to the FLT, which is quite easy to do.  It would look like the
 MB> Unix "mount" command.
Yes. Perhaps I should extend my DevCon subscription to DevCon+DDK. Or could
anyone make the LOCKDRV-sources available for me ? (FTP preferable)
     __
     |_)
cu/2 |atrick
[Team  OS/2]

---