On Thu, 26 Aug 2004, David S. Miller wrote:
On Thu, 26 Aug 2004 15:38:49 +0200 (CEST) Henrik Nordstrom <hno@marasystems.com> wrote:
From reading the kernel source it seems the local port assignment function has much to ask, only assigning a source port unless that source port is not used by any connection on the system irregardless of the source IP. Is this really true?
Yes, currently it is.
Do you know if anyone is looking into making the local endpoint assignment more friendly?
I tried to look into doing something about this, but it seems to be limited by design and not only implementation with the port and address assigned independently by different methods in the protocol(s).
Issues where this can be improved:
*) Ability to use more than one source IP for the same route.
*) Local endpoint assignment assigning ip:port as one, not in steps, allowing it to use the multiple IP addresses from above in an efficient manner, or at a minimum limited to the source IP of the current route not globally (requires source IP to be assigned first).
*) In case of connect() the local endpoint assignment only needs to guarantee the connection tuple is unique. The local enpoint alone does not need to be unique as the same ip:port can be used for connections to different remote endpoints. This would considerably lift the connection limit for a single IP address in most real-world situations (but not in my current synthetic workloads however..).
Regards Henrik - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
