Skip to contentSkip to conversation

Router dropping packets? - SOLVED

LARTC John Philips 2 participants 2 messages Mar 7, 2007 – Mar 30, 2007

lore
Share & tools
lore
#0 lore raw

Guys,

I called my DSL provider and it turns out they limit the number of simultaneous "flows" you can have. I guess that means active TCP connections. Their limit is 1500 concurrent flows, and when the tech looked at it we had 1450 active.

I presume all these flows are from P2P users, so I'm going to try using the connlimit iptables extension to prevent individual users from having more than 50 or so connections.

Quoted 52 lines · --- John Philips

Hey guys,

I have several Linux routers in place at high-usage locations (student apartment complexes). I'm having trouble with some of the routers which use 6Mbit DSL lines as their Internet feed. The routers use PPPoE and perform NAT.

During peak usage periods, the routers are dropping alot of packets. I'm lead to believe this is because
there are too many active connections.

For example, when I ping the WAN IP address of one of
the routers from a remote location, I may start getting replies immediately. But during peak periods,
the first several pings usually time out and then they
just start responding. Sometimes they start responding on the 4th ping, sometimes the 12th, etc.,
it's pretty random.

I searched the web and tried increasing my gc_cache settings, but it didn't make a difference.

echo 512 > /proc/sys/net/ipv4/neigh/default/gc_thresh1 echo 2048 > /proc/sys/net/ipv4/neigh/default/gc_thresh2 echo 4096 > /proc/sys/net/ipv4/neigh/default/gc_thresh3

The other notable difference is that the conntrack tables are much larger than normal.

wc -l /proc/net/ip_conntrack returns >19000 on the routers experiencing packet loss while virtually all of the other routers (not having this issue) have less
than 5000 entries in ip_conntrack. I tried increasing
ip_conntrack_max in /proc, setting it to 65536 - didn't make a difference.

Are there any other /proc settings I should change to
improve performance? Any tips on analyzing the ip_conntrack data to find oddities?

Quoted 12 lines

FYI I'm using kernel 2.4.25. I'd rather not upgrade to 2.6 since doing so in the past has introduced more
problems!

Thanks.

____________________________________________________________________________________

Quoted 7 lines

No need to miss a message. Get email on-the-go with Yahoo! Mail for Mobile. Get started.
http://mobile.yahoo.com/mail
_______________________________________________
LARTC mailing list
[email protected]

http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Quoted 1 line

____________________________________________________________________________________
Don't get soaked. Take a quick peek at the forecast with the Yahoo! Search weather shortcut.
http://tools.search.yahoo.com/shortcuts/#loc_weather
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#1 lore raw
Quoted 5 lines · John Philips

Guys,

I called my DSL provider and it turns out they limit the number of simultaneous "flows" you can have. I guess that means active TCP connections.

But you saw problems with icmp...., but as Corey said you may have Qos problems aswell

Their limit

Quoted 2 lines

is 1500 concurrent flows, and when the tech looked at it we had 1450 active.

Ewww that's horrible - I can eat 2k conntracks all by myself.

Andy.
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc