Actually still triggers with NMBCLUSTERS at 65536, after roughly 1h, 10m, 38s. 
The network hang is for about 7 minutes.

WARNING: mclpool limit reached; increase NMBCLUSTERS
WARNING: mclpool limit reached; increase NMBCLUSTERS

How high can I keep pushing NMBCLUSTERS safely on a 1Gb machine?

load averages:  2.58,  1.47,  1.65                                     09:57:48
30 processes:  29 sleeping, 1 on processor
CPU states: 10.0% user,  7.0% nice,  5.5% system,  9.0% interrupt, 68.7% idle
Memory: 583M Act, 291M Inact, 324K Wired, 5904K Exec, 854M File, 384K Free
Swap: 2048M Total, 2048M Free

Lund

Jorgen Lundman wrote:
> 
> NetBSD  3.0 NetBSD 3.0 (netbsd) #0: Tue Dec 27 12:49:48 JST 2005 
> root@netbsd:/usr/src/netbsd-3.0/sys/arch/i386/compile/netbsd i386
> 
> I was hitting this bug quite frequently:
> 
> http://www.netbsd.org/cgi-bin/query-pr-single.pl?number=32287
> 
> 
> 1179 ?     DWs  100:55.29 ./lundftpd
> 
>  1179 root     -22    0  6408K    4K mclpl    100:55  0.00%  0.00%
>  <lundftpd>
> 
> sysctl kern.somaxkva
> kern.somaxkva = 16777216
> 
> (gdb) p nmbclusters
> $1 = 32768
> 
> NIC down/up did nothing, increasing sokva did not "unstick" it. Only 
> reboot fixed the issue.
> 
> 
> I tried to compile -current for the fixes mentioned, but -current is not 
> easy to compile from 3.0.x at the moment. Seems to want a new share/mk, 
> gcc v4 ("--no-pointer-sign & --no-attributes" are pulled in even though 
> it does check for GCC_VERSION > 4).
> 
> So instead I patched in the new versions of uvm_map.[ch] uvm_km.c.  I 
> did not patch the "src/sys/kern/subr_pool.c" file. Perhaps I should?
> 
> 
> Anyway, we have managed to run twice as long as previous record with new 
> patches, so they appear to make things better.
> 
> Now when the bug would have triggered I merely get:
> 
> WARNING: mclpool limit reached; increase NMBCLUSTERS WARNING: mclpool 
> limit reached; increase NMBCLUSTERS
> WARNING: mclpool limit reached; increase NMBCLUSTERS
> 
> .. and a short hang.
> 
> I will go to 65536 and try.
> 
> But apart from that, just reporting that the patches at least appears to 
> have lifted the need to reboot every day.
> 
> Lund
> 
> 
> 

-- 
Jorgen Lundman       | <lundman@lundman.net>
Unix Administrator   | +81 (0)3 -5456-2687 ext 1017 (work)
Shibuya-ku, Tokyo    | +81 (0)90-5578-8500          (cell)
Japan                | +81 (0)3 -3375-1767          (home)