I got a response to my SOMAXCONN question that might be of
use to others..
If it works (I might know tomorrow) it'd be worth documenting this.
Mark Thomas wrote:
>
> Rob Hartill (hartill@lanl.gov) wrote:
> > Given that changing SOMAXCONN in /usr/sys/sys/socket.h
> > seemed to have no affect on our SunOs Sparc 5 running a
> > HTTP server, and that "SOMAXCONN" doesn't seem to be referenced
> > in any of the source files we have to build a new kernel, does
> > anyone know how to or can point me at information that'll
> > let me increase the SOMAXCONN setting?
>
> I ran into the same problem. I run a 110 MHz SPARC-5
> clone running 4.1.4, and periodically the web server would
> completely stop accepting connections on port 80 for anywhere from
> several second to several minute periods. There was a correlation
> between these times, and netstat reporting five or six connections
> persisting in the SYN_RCVD state.
>
> Increasing SOMAXCONN from 0x5 to 0x42 solved the problem, as now even
> with peaks of 10 to 15 connections in SYN_RCVD, new connections are never
> blocked. Currently this machine is serving about 300,000
> web connections per day using that harvest cache httpd accelerator;
> performance is great and there are no more stalls.
>
> SOMAXCONN is compiled into uipc_socket.o by Sun, so the only way
> to change it is to modify uipc_socket.o, then build your kernel.
>
> My binary mods to uipc_socket.o for sun4m are included below,
> but use them at your own risk. Also, I am running the uipc_socket.o
> module from patch 102426-03. I have not noticed any problems, but
> if anyone else does this, I would be interested in hearing about
> your results.
>
> % od -h /sys/sun4m/OBJ/uipc_socket.o
>
> before:
> 0000700 d456 2002 9412 a002 d436 2002 8090 0019
> 0000720 3680 0004 80a6 6005 b210 2000 80a6 6005
> ^^ ^^
> 0000740 2680 0004 f236 2020 b210 2005 f236 2020
>
> after:
> 0000700 d456 2002 9412 a002 d436 2002 8090 0019
> 0000720 3680 0004 80a6 6042 b210 2000 80a6 6042
> ^^ ^^
> 0000740 2680 0004 f236 2020 b210 2005 f236 2020
>
> -Mark
>
> --
> Mark G. Thomas (Mark@Misty.com -- http://www.misty.com)
use to others..
If it works (I might know tomorrow) it'd be worth documenting this.
Mark Thomas wrote:
>
> Rob Hartill (hartill@lanl.gov) wrote:
> > Given that changing SOMAXCONN in /usr/sys/sys/socket.h
> > seemed to have no affect on our SunOs Sparc 5 running a
> > HTTP server, and that "SOMAXCONN" doesn't seem to be referenced
> > in any of the source files we have to build a new kernel, does
> > anyone know how to or can point me at information that'll
> > let me increase the SOMAXCONN setting?
>
> I ran into the same problem. I run a 110 MHz SPARC-5
> clone running 4.1.4, and periodically the web server would
> completely stop accepting connections on port 80 for anywhere from
> several second to several minute periods. There was a correlation
> between these times, and netstat reporting five or six connections
> persisting in the SYN_RCVD state.
>
> Increasing SOMAXCONN from 0x5 to 0x42 solved the problem, as now even
> with peaks of 10 to 15 connections in SYN_RCVD, new connections are never
> blocked. Currently this machine is serving about 300,000
> web connections per day using that harvest cache httpd accelerator;
> performance is great and there are no more stalls.
>
> SOMAXCONN is compiled into uipc_socket.o by Sun, so the only way
> to change it is to modify uipc_socket.o, then build your kernel.
>
> My binary mods to uipc_socket.o for sun4m are included below,
> but use them at your own risk. Also, I am running the uipc_socket.o
> module from patch 102426-03. I have not noticed any problems, but
> if anyone else does this, I would be interested in hearing about
> your results.
>
> % od -h /sys/sun4m/OBJ/uipc_socket.o
>
> before:
> 0000700 d456 2002 9412 a002 d436 2002 8090 0019
> 0000720 3680 0004 80a6 6005 b210 2000 80a6 6005
> ^^ ^^
> 0000740 2680 0004 f236 2020 b210 2005 f236 2020
>
> after:
> 0000700 d456 2002 9412 a002 d436 2002 8090 0019
> 0000720 3680 0004 80a6 6042 b210 2000 80a6 6042
> ^^ ^^
> 0000740 2680 0004 f236 2020 b210 2005 f236 2020
>
> -Mark
>
> --
> Mark G. Thomas (Mark@Misty.com -- http://www.misty.com)