Problems with load balancing, EIGRP

[bej]

I have 4 2501 routers and 2 computers. PC 1 is using Win XP and PC 2 Win 98. All routers are running IOS 2500-I-L, ver 12.1(2)T. When i send data from PC 2 to PC 1, traffic will not load balance across the 2 available paths. See diagram and configuration.

Show outputs on LabD

LabD#sh ip route

N1 - OSPF NSSA external type 1, N2 - OSPF NSSA external type 2

E1 - OSPF external type 1, E2 - OSPF external type 2, E - EGP

i - IS-IS, L1 - IS-IS level-1, L2 - IS-IS level-2, ia - IS-IS inter area

* - candidate default, U - per-user static route, o - ODR

P - periodic downloaded static route

Gateway of last resort is not set

C 192.168.30.0/24 is directly connected, Loopback0

C 192.168.60.0/24 is directly connected, Loopback3

C 192.168.10.0/24 is directly connected, Serial1

C 192.168.40.0/24 is directly connected, Loopback1

172.16.0.0/24 is subnetted, 3 subnets

D 172.16.205.0 [90/4096000] via 192.168.10.15, 01:05:12, Serial1

[90/4224000] via 172.16.56.44, 01:05:12, Serial0

C 172.16.56.0 is directly connected, Serial0

D 172.16.31.0 [90/3584000] via 192.168.10.15, 01:05:12, Serial1

C 192.168.20.0/24 is directly connected, Ethernet0

10.0.0.0/24 is subnetted, 1 subnets

D 10.10.10.0 [90/3609600] via 192.168.10.15, 01:05:13, Serial1

[90/4249600] via 172.16.56.44, 01:05:13, Serial0

C 192.168.50.0/24 is directly connected, Loopback2

D 192.168.0.0/18 is a summary, 01:05:16, Null0

LabD#sh ip eigrp topo 10.10.10.0 255.255.255.0

IP-EIGRP topology entry for 10.10.10.0/24

State is Passive, Query origin flag is 1, 1 Successor(s), FD is 3609600

Routing Descriptor Blocks:

192.168.10.15 (Serial1), from 192.168.10.15, Send flag is 0x0

Composite metric is (3609600/3097600), Route is Internal

Vector metric:

Minimum bandwidth is 1000 Kbit

Total delay is 41000 microseconds

Reliability is 255/255

Load is 1/255

Minimum MTU is 1500

Hop count is 2

172.16.56.44 (Serial0), from 172.16.56.44, Send flag is 0x0

Composite metric is (4249600/1817600), Route is Internal

Vector metric:

Minimum bandwidth is 800 Kbit

Total delay is 41000 microseconds

Reliability is 255/255

Load is 1/255

Minimum MTU is 1500

Hop count is 2

LabD#trace 10.10.10.1

Type escape sequence to abort.

Tracing the route to 10.10.10.1

1 192.168.10.15 4 msec

172.16.56.44 4 msec

192.168.10.15 4 msec

2 172.16.205.1 8 msec

172.16.31.1 8 msec

172.16.205.1 8 msec

3 10.10.10.1 8 msec 4 msec 4 msec

Traceroute from PC 2:

tracert 10.10.10.1

1 3ms 3ms 3ms 192.168.20.179

2 4ms 4ms 5ms 172.16.56.44

3 6ms 6ms 6ms 172.16.205.1

4 5ms 5ms 5ms 10.10.10.1

[bej]

Here is the diagram

[bej]

Here is the diagram

[Georg Pauwen]

Hello,

from your trace, you can see that the topology with the two paths is configured correctly. Possibly, you load-balancing scheme is per-destination. By default, on most Cisco routers, fast switching is enabled under interfaces. This is a demand caching scheme that does per-destination load-balancing. To set per-packet load-balancing, enable process switching (or disable fast switching) using the following commands:

Router# config t

Router(config)# interface Ethernet 0

Router(config-if)# no ip route-cache

Router(config-if)# ^Z

With per-packet load-balancing, packets are evenly distributed among the available links.

Can you try this on your router and see if it makes a difference ?

Regards,

Georg

www.solutionfinders.nl

[Harold Ritter]

Take most be taken when using per-packet loadsharing. I have seen more issues caused by it than performance improvement. This is mostly related with TCP retransmissions that could be caused by out of sequence packets.

Hope this helps,

[bej]

Thanks for your replies. I tried to use per-packet loadsharing, it did not make a difference. I am using Pcattcp to measure throughput from PC 2 to PC 1. With both interfaces on LabD up i get about 116KBytes/s, with interface serial 1 shutdown i get approx. 90 KBytes/s throughput. With load balancing i would expect approx. 1800kbits throughput to PC 1

[Harold Ritter]

This is exactly what I was referring to in my previous posting. When using TCP with per-packet loadsharing, you can't expect the throughput to double just because you have twice the bandwidth. If you use a sniffer trace, you will probably see alot of TCP retransmissions, which is not too good performancewise.

Hope this helps,