Bob Beck wrote:
Try this.
We used to do this, but perhaps prematurely. I've now killed several of
the nfs bugs this used to tickle.
This lets you make more effective use of your larger buffer cache, if
you are using it for lots of small files (like /usr/src, and the like)
since you can cache more than the default of about 5000 odd files with
the old code since you'd run out of vnodes and start recycling them.
Please let me know if you see any problems, or if you notice
speedups/slowdowns with this on things you like to do.
I have a one-gig machine which have no issues at all, it gets something
like 57k vnodes and stays there, having been pushing it for a day
without issues, then I tried it on my 2G ram laptop, got somewhere over
107k vnodes and it hung. I was in X11 so no cookie for me from ddb>, but
I'll redo the tests later tonight when I dont need my machine for a 60G
fsck session or two.
Both run bufcachepercent at 90.
dmesg first machine
http://people.su.se/~jj/obsd/dmesg.1g
dmesg second machine
http://people.su.se/~jj/obsd/dmesg.2g
I cant say if the extra gig ram is the "obvious" killer here, the first
machine is a desktop and has scsi, the second a laptop with sata but I
gather the desktop 1G ram makes the kernel limit to 50k+ vnodes instead
of 100k+ and perhaps that makes the first machine survive it.
