Hi,

I recently started using a ccd-drive as my /usr, a I wondered if there
is a good way to guess a "reasonable" interleave factor (as this is my
/usr-drive and each attempt basically requires a newfs/restore-cycle, 
I try to avoid trial-and-error-methods). 

For the time being, I used and interleave of 32 as in one of the 
examples in the ccdconfig(8) manpage. While this works, it is a bit
slower than I expected:

| zork.north.de:/home/gizmo# dd if=/dev/ccd0d of=/dev/null bs=65536
| 1869+0 records in
| 1869+0 records out
| 122486784 bytes transferred in 42 secs (2916352 bytes/sec)
|
| zork.north.de:/home/gizmo# dd if=/dev/rccd0d of=/dev/null bs=65536
| 3155+0 records in
| 3155+0 records out
| 206766080 bytes transferred in 34 secs (6081355 bytes/sec)

I had hoped for a bit more than 8 MB/sec for the second example (and
yes, dd is not exactly a real-world test. :-) )

The decision of a good interleave factor is made a bit more
complicated, because the drives are similar, but not identical:

sd1 at scsibus0 targ 1 lun 0: <SEAGATE, ST32430N, 0510> SCSI2 0/direct fixed
sd1: 2049MB, 3992 cyl, 9 head, 116 sec, 512 bytes/sec
sd5 at scsibus0 targ 5 lun 0: <SEAGATE, ST32155N, 0594> SCSI2 0/direct fixed
sd5: 2049MB, 4177 cyl, 8 head, 125 sec, 512 bytes/sec

(sd1 is an older Fast-SCSI Hawk, while sd5 is a new Ultra-SCSI Hawk).
I did, however, disklabel sd5 in a way to make it as similar to sd1 as
possible:

116 sec, 9 head, 4000 cyl

I also made the 'e'-partition of the drives  exactly the same size
(both drives are exclusively used for NetBSD). At this time, both
drives are at the same SCSI-bus (Adaptec 2940), I intend to get an
additional Adaptec 2940U soon (for sd5). (oh yes, this is
NetBSD/i386-1.2.1, it it matters).

Finally, here is my /etc/ccd.conf:

| # ccd           ileave  flags   component devices
| ccd0            32      none    /dev/sd5e /dev/sd1e


So, to summarize: could anyone having more experience in setting up
ccd-drives suggest a good interlave value for my setup?


Thanks in advance,
	Klaus