Tunable limitting number of established BSD sockets

[Ralph Grothe]

Hi,

this request is linked to my previous thread

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=246754

But I thought to maybe draw more attention to if I started a new thread.

While searching the knowledge base for limitting factors for the inetd super daemon I came accross this thread

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=56669

There Mark and Todd stress that the kernel tunable maxusers is one limitting factor.

No wonder I now must conclude.
This box is almost always scraping at the the set boundary by maxusers.

# kmtune -q maxusers
Parameter Value
===============================================================================
maxusers 1200

# netstat -an -f inet|grep -c ESTAB
1711


Not to mention other tunables that constitute bounds for files in the Unix sense (viz. comprising sockets, fifos, locks, files etc.)

# kmtune -q maxusers -q nproc -q maxfiles -q maxfiles_lim -q nfile -q nflocks
Parameter Value
===============================================================================
maxusers 1200
nproc ((MAXUSERS*3)+64)
maxfiles 1024
maxfiles_lim 2048
nfile (8*NPROC+2048)
nflocks (NPROC)


Would you agree that increasing maxusers could be the key?

[Stefan Farrelly]

Youre certainly very tight on maxusers. But this parameter is there to stop too many users getting on because your server doesnt have enough resources for more users.

If you have some spare ram, cpu etc. then I would certainly bump maxusers up, if not then I wouldnt unless you can increase your server spec (more hardware).