Kernel TCP tuning

Mauro Souza thoriumbr-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org
Fri Feb 8 23:47:40 UTC 2013


You can try setting a lower limit, like 5 ports, and using netcat to open 4
connections. Then use strace/ptrace on a curl or wget and see what happens.
On Feb 8, 2013 8:45 PM, "Ben Walton" <bdwalton-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org> wrote:

> What condition do you need to detect? Are you looking to solve/monitor an
> application or system problem? I expect the kernel would log something but
> the source our experimentation are the oracles for this as I've never tried
> it.
>
> Also, why would you do this? (For my own curiosity...)
>
> Thanks
> -Ben
> On Feb 8, 2013 4:21 PM, "Tyler Aviss" <tjaviss-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org> wrote:
>
>> My understanding is that the following setting basically sets the range
>> of high "local ports" used for outgoing connections
>>
>> Either in sysctl.conf
>> /proc/sys/net/ipv4/ip_local_port_range = LOW HIGH
>>
>> This can also be set by using sysctl or echoing to proc (but lost on
>> reboot).
>>
>>
>>
>> My question is, what would be a symptom of exhausting the port pool?
>> Say if you have a port range of 10001 15000, and you attempt connection
>> #5001... what happens?
>>
>> Is there something visible in dmesg/kern.log/syslog that indicates an
>> error opening ports? Do you see the port opening/connection attempt time
>> out, or just block?
>>
>> I know if you start nipping at the limits of your max filehandles, an
>> kernel error is rather apparent, but I'm not sure what happens for ports.
>>
>> As most people prefer to leave this at the default, there isn't a lot of
>> information other than how to set the port range
>>
>>
>>
>>
>> --
>> Tyler Aviss
>> Systems Support
>> LPIC/LPIC-2/DCTS/CLA
>>
>> "Computers don't make mistakes. They can, however, execute those provided
>> to them very quickly"
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://gtalug.org/pipermail/legacy/attachments/20130208/53f056af/attachment.html>


More information about the Legacy mailing list