Folks,

I've got a NetBSD/pmax machine (a 5000/240 with 24MB of RAM) running
NetBSD 1.3.  It's got 7x 2GB RZ28 SCSI disks - one "boot" disk and the
other six striped on a ccd.  It's only purpose is to NFS export the
ccd filesystem; this filesystem holds ftp area for our group, and is
exported read/write to the machine that handles the actual ftping and
read-only to the rest of the world (inside our company).

Quite often (a couple of times a day), I get the following panic:

	Apr 21 00:55:12 fhfss /netbsd: panic: malloc: out of space in kmem_map
	Apr 21 00:55:12 fhfss /netbsd: syncing disks... 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 giving up

and lots of:

	Apr 20 22:00:19 fhfss /netbsd: mb_map full
	Apr 20 22:01:19 fhfss /netbsd: mb_map full
	Apr 20 22:02:49 fhfss /netbsd: mb_map full
	Apr 20 22:13:19 fhfss last message repeated 8 times
	Apr 20 22:22:49 fhfss last message repeated 7 times
	Apr 20 22:33:19 fhfss last message repeated 8 times
	Apr 20 22:42:49 fhfss last message repeated 7 times
	Apr 20 22:53:19 fhfss last message repeated 8 times
	Apr 20 23:02:49 fhfss last message repeated 7 times
	Apr 20 23:13:19 fhfss last message repeated 8 times
	Apr 20 23:22:49 fhfss last message repeated 7 times
	Apr 20 23:33:19 fhfss last message repeated 8 times
	Apr 20 23:42:49 fhfss last message repeated 7 times
	Apr 20 23:53:19 fhfss last message repeated 8 times
	Apr 21 00:02:49 fhfss last message repeated 7 times
	Apr 21 00:13:19 fhfss last message repeated 8 times
	Apr 21 00:22:49 fhfss last message repeated 7 times
	Apr 21 00:33:19 fhfss last message repeated 8 times
	Apr 21 00:42:49 fhfss last message repeated 7 times

which occured (in the case) immediately before the above panic.

Also, the clients frequently get "NFS server not responding".  When this
happens, I can rlogin to the machine without problems, but "netstat -m"
shows an increasing number of mbufs being used.  For example:


	fhfss:~ 28> netstat -m ; sleep 60 ; echo "" ; netstat -m ; uptime
	563 mbufs in use:
		375 mbufs allocated to data
		188 mbufs allocated to socket names and addresses
	1/6 mapped pages in use
	82 Kbytes allocated to network (87% in use)
	0 requests for memory denied
	0 requests for memory delayed
	0 calls to protocol drain routines

	590 mbufs in use:
		393 mbufs allocated to data
		197 mbufs allocated to socket names and addresses
	1/6 mapped pages in use
	85 Kbytes allocated to network (88% in use)
	0 requests for memory denied
	0 requests for memory delayed
	0 calls to protocol drain routines
	 8:43AM  up 25 mins, 1 user, load averages: 0.07, 0.10, 0.15


The basic changes I've made from the GENERIC kernel are:

	options       "NMBCLUSTERS=4096"      # extra mbuf clusters
	options       CCDNBUF=16              # number of buffers per ccd device


Anyone got any ideas?  If I get no answers like "you forgot to do this",
I'll send-pr this problem.

Simon.