|
Hello Julian,
> I just uploaded fastcsum-1.0.8-2.2.19-1.diff in
> http://www.linuxvirtualserver.org/~julian/fastcsum-1.0.8-2.2.19-1.diff
Verrrry cool!
> This patch includes the same optimizations we can see in the
> development IPVS 0.9.4 for 2.4, i.e. incremental checksum updates for
> all LVS-NAT connections and in the 2.2 case even for the masqueraded
> (non-LVS) ones.
Some beginner question to your patch:
ip_masq_check_tcpudp():
+#ifdef CONFIG_IP_MASQ_DEBUG
+ if (ip_masq_get_debug_level() > 3) {
+ skb->ip_summed = CHECKSUM_NONE;
+ }
+#endif
Why do you set CHECKSUM_NONE if CONFIG_IP_MASQ_DEBUG is set? What
happens
if when before calling ip_masq_check_tcpudp() we have a wrong checksum?
> This is a 1.0.9 candidate, so your test results are welcome. The
> reason to announce it before 1.0.9 is to allow some LVS users to test it
> and to report their results. The interesting part is whether there is a
> speed improvement for the NAT processing. Of course, such tests can
> be performed with CONFIG_IP_ROUTER defined (if used from your NIC at all).
> If everything is ok, we will see a new 1.0.9 release (the last was from
> May).
Don't you think we should wait for 2.2.20 to come out? Alan should have
released it since 2 weeks already. I wonder if he was flushed down the
toiled or if he's so busy trying to divert 2.4.x-ac series as much as
possible from the Linus-series. 2.2.20 has tons of TCP stack fixes.
YMMV, but still best regards,
Roberto Nibali, ratz
|
