Stoned koala bears drooled eucalyptus spit in awe as Dave McGuire exclaimed:


>    Do you know offhand if a 1.6L kernel coexist nicely with a 1.6
> userland?

I ran a 1.6L kernel from the -current snapshot with a 1.6 userland and 
didn't have any problems other than....

the 1.6 mount acts kinda funny at boot.  When it first tries to remount / 
read-write, it complains something about "realpath: / bad address" then I 
get all kinds of errors when it tries chmod'ing all the /dev/tty entries. 
Then sshd fails to initially start because /dev/random apparently isn't 
"primed" at that point[1].  What's odd is that adding a line to run sshd 
from rc.local works fine.

I don't have any of these problems booting a single-processor kernel.

>    Is it stable otherwise?  (the system I'm thinking about is a
> production machine...the added performance would be a big benefit for me)

For me, other than the NFS hangs, it seems to work fine. Mine is hardly a 
production machine, though, so YMMV

>> FTP speed is much better I had previously reported 2.19
>> megabyte/second, tonight I got 5.2 megabyte/second.  There were still
>> several pauses for no apparent reason during the download, though,
>> some lasting several seconds.
>
>    Wow, that's quite a difference in throughput.  I wonder what's
> happening with those delays; I haven't seen anything like that on my 1.6
> system.

I'm thinking about putting some disks in a RAIDFrame array so I can do some 
performance testing on a disk not on the DAC960 to see if it's isolated to 
the RAID controller.  At first i was thinking that maybe the kernel was 
flushing some sort of buffer and causing the pauses, but if that was the 
cause wouldn't it be doing it at fairly regular intervals?  Plus, this is 
happening during disk reads, not writes[2].

-Jeff

[1] Incidentally I still have the sshd problem using a "mount" from 1.6L.
[2] Although, it may occur during writes, too.  I've not tested.

--
Jeff Workman | jworkman@pimpworks.org | http://www.pimpworks.org