i fixed these load-related issues by 3 things:
1) set per-process file descripter limits to 8192
2) set system tunables net.core.rmem_max and net.core.wmem_max to 16777216
3) set the third figure in net.ipv4.tcp_rmem and net.ipv4.tcp_wmem to 16777216
these were done on RHEL6.
pieter, can you add these to the tuning wiki? thanks
andrew
On Jun 14, 2011, at 10:15 AM, Pieter Hintjens wrote:
> On Tue, Jun 14, 2011 at 3:01 PM, Andrew Hume <[email protected]> wrote:
>
>> Assertion failed: new_sndbuf > old_sndbuf (mailbox.cpp:183)
>
> The assertion is annoyingly useless, IMO the code should at least say
> something like "Insufficient socket buffer space available" before
> dying.
>
> The problem is that the OS isn't providing enough buffer space for the
> socketpair used for internal mailbox communications. Linux does set
> its limits automatically but RHEL6 might be not doing this. Here's a
> page that may help:
> http://www.psc.edu/networking/projects/tcptune/#Linux
>
> The problem is more typically seen on BSD systems (Mac OS/X and
> FreeBSD): http://www.zeromq.org/docs:tuning-zeromq
>
> -Pieter
> _______________________________________________
> zeromq-dev mailing list
> [email protected]
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
------------------
Andrew Hume (best -> Telework) +1 623-551-2845
[email protected] (Work) +1 973-236-2014
AT&T Labs - Research; member of USENIX and LOPSA
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev
