Re: major network performance difference between 2.4 and 2.6.2-rc2

[Bill Davidsen]

Jim Faulkner wrote:

> Thanks to Andrew's suggestion of profiling my kernel, I've figured out
> what is happening here.  It is my fault, it is not a bug.
>
> I use this iptables script generator:
> http://ftp.berlios.de/pub/mldonkey/pango/goodies/ipblacklist_convert
> in combination with this blacklist:
> http://www.peerguardian.net/pgipdb/guarding.p2p
>
> I had already modified the script so everything on my LAN interface was
> accepted, however I didn't realize that the scipt was using "-I INPUT 1"
> for all of its blacklist rules.  iptables was going through around 5300
> rules for each and every packet that came in through my LAN interface,
> which is definately not what I intended.
>
> I fixed my firewall script, and my LAN throughput is back up at 10
> megabytes per second, with nowhere near the load.

This does point out an issue, as a 2.7 enhancement it would be really 
useful to have a better way to handle a large number of rules, when what 
you want is one rule applied to many IP values. I ran into this when 
fighting a DDoS attack, and by the time I got the attack stopped, even 
only dropping or rejecting --syn packets I had most of a CPU in system 
time running ~10k rules.

I wrote a perl script to break it into multiple level tables, but it was 
still pretty slow and uglier than a hedgehog's rectum.

What would be nice is some kind of table approach, hash or tree, which 
allows operations to be matches against all of the IPs in a group, and 
obviously to add/delete entries. I think for simplicity individual IPs 
rather than CIDR blocks are desirable.

In any case, if a network person is looking for something really neat 
for 2.7, blactlists of various types are getting more common, and an 
efficient solution would be good.


--
bill davidsen <davidsen@xxxxxxx>
  CTO TMR Associates, Inc
  Doing interesting things with small computers since 1979
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/