Is there a MAX TCP/UDP CONNECTIONS limit in Kernel?

Is there a MAX TCP/UDP CONNECTIONS limit in Kernel?

[Fernando Netto]

Hi,

I´m having problems with concurrent TCP/UDP connections in my machine and
wondered if there is a limit in Kernel of "how many connections can be open
simoultaneously".

If someone know something about limitations and how to tune it up about this
matter, please don´t forget to put my address in CC as I´m not a signer of
this list.

Thank you,


Fernando Netto

Re: Is there a MAX TCP/UDP CONNECTIONS limit in Kernel?

[Sven Koch]

On Thu, 1 Nov 2001, Fernando Netto wrote:

> I´m having problems with concurrent TCP/UDP connections in my machine and
> wondered if there is a limit in Kernel of "how many connections can be open
> simoultaneously".

Are you talking of inbound, or outbound connections?

Inbound-connections are mostly limited by the available filehandles and by
how much server-processes your box can handle.

Outbound-connections are limited by the local portrange, changeable
in /proc/sys/net/ipv4/ip_local_port_range
(ran into this on one of my proxy servers, having thousands of connections
in the state CLOSING, TIME_WAIT and LAST_ACK - after
echo "1024 16383" >/proc/sys/net/ipv4/ip_local_port_range the box at
least stays working)

> If someone know something about limitations and how to tune it up about this
> matter, please don´t forget to put my address in CC as I´m not a signer of
> this list.

c'ya
sven

-- 

The Internet treats censorship as a routing problem, and routes around it.
(John Gilmore on http://www.cygnus.com/~gnu/)

Re: Is there a MAX TCP/UDP CONNECTIONS limit in Kernel?

[Till Immanuel Patzschke]

... there may be some sort of "implicit" limit resulting from the kernel memory.
I've run into the same problem but a fix is easy.  Get the 3.5G patch from
Andrea's patch series (the -aa one), and modify it to NOT have 3.5GB of user
memory but 2.5 or 2 GB for example. Worked fine for me...
Cheers,

Immanuel


Sven Koch wrote:

> On Thu, 1 Nov 2001, Fernando Netto wrote:
>
> > I´m having problems with concurrent TCP/UDP connections in my machine and
> > wondered if there is a limit in Kernel of "how many connections can be open
> > simoultaneously".
>
> Are you talking of inbound, or outbound connections?
>
> Inbound-connections are mostly limited by the available filehandles and by
> how much server-processes your box can handle.
>
> Outbound-connections are limited by the local portrange, changeable
> in /proc/sys/net/ipv4/ip_local_port_range
> (ran into this on one of my proxy servers, having thousands of connections
> in the state CLOSING, TIME_WAIT and LAST_ACK - after
> echo "1024 16383" >/proc/sys/net/ipv4/ip_local_port_range the box at
> least stays working)
>
> > If someone know something about limitations and how to tune it up about this
> > matter, please don´t forget to put my address in CC as I´m not a signer of
> > this list.
>
> c'ya
> sven
>
> --
>
> The Internet treats censorship as a routing problem, and routes around it.
> (John Gilmore on http://www.cygnus.com/~gnu/)
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

--
Till Immanuel Patzschke                 mailto: tip@internetwork-ag.de
interNetwork AG                         Phone:  +49-(0)611-1731-121
Bierstadter Str. 7                      Fax:    +49-(0)611-1731-31
D-65189 Wiesbaden                       Web:    http://www.internetwork-ag.de

Re: Is there a MAX TCP/UDP CONNECTIONS limit in Kernel?

[David S. Miller]

   From: Sven Koch <haegar@sdinet.de>
   Date: Thu, 1 Nov 2001 19:37:31 +0100 (CET)
   
   Outbound-connections are limited by the local portrange, changeable
   in /proc/sys/net/ipv4/ip_local_port_range
   (ran into this on one of my proxy servers, having thousands of connections
   in the state CLOSING, TIME_WAIT and LAST_ACK - after
   echo "1024 16383" >/proc/sys/net/ipv4/ip_local_port_range the box at
   least stays working)

In current 2.4.14-preX, this is not true anymore.  It is limited
by something approximating "local port range X number of unique
destination IP addresses" because if the remote address is unique
we allow multiple local port binds to occur on the same local port.

Franks a lot,
David S. Miller
davem@redhat.com