load balancing

RouterOS General
1

I have a setup with 2 ethernet WANs each with a 2mbps connection.

I have load balancing per session working OK with NAT but with a few glitches.

1/ Every now and then browsing stalls or a page takes too long to appear.

2/ Large downloads stop after a few mins but then start again after a min.

Obviously there is some problem with the routing as the route changes to each WAN.

Has anybody overcome this or can offer some tips?

Thanks in advance.

2

What are your connection tracking timeouts set to? Are they default? Maybe your tcp sessions are timing out too quickly.

When you say ‘the route changes to each WAN’ - what do you mean? If you are using ECMP there should be multiple default gateways active all the time and no switching should happen. Is one wan interface going inactive?

Also - single or multiple physical interfaces?

Sam

3

Hi,

Yes all my conntrack settings are default.

QUOTE"When you say ‘the route changes to each WAN’ - what do you mean? If you are using ECMP there should be multiple default gateways active all the time and no switching should happen. Is one wan interface going inactive?

Also - single or multiple physical interfaces? "

I just mean that it seems as if the client pc has a problem when the masquaraded IP changes, but thats only theory?

Both internet connections are on different physical ethernet interfaces and neither one goes down.

I don’t see any reference to ECMP i’m using 2.8 if thats any help.

Thanks again.

4

can you post the results of an /ip route print ?

Sam

5

0 S ;;; Node4 Gate
0.0.0.0/0 r 83..184. 1 ether1
1 XS ;;; Zen Gate
0.0.0.0/0 u 82..98.2* 1 (unknown)
2 XS ;;; Load balance gates
0.0.0.0/0 u 83..184. 1 (unknown)
u 82.1.8.2 (unknown)
3 S 83..184.250/32 r 83..184.* 1 ether1
4 S 83..160.10/32 r 83..184.** 1 ether1
5 S 83..160.21/32 r 83..184.** 1 ether1
6 S 2**.23.8.1/32 r 82..98.24 1 ether4
7 DC 192.168.2.0/24 r 0.0.0.0 0 ether3
8 DC 192.168.1.0/24 r 0.0.0.0 0 ether2
9 DC 83.6.*.32/28 r 0.0.0.0 0 ether1
10 DC .71..240/29 r 0.0.0.0 0 ether4

Funny thing is I tried policy routing instead only p2p ports and newsgroups to the cheaper line and the problem shows up there too when the client PC has been using one gateway and then browses a site which is in the other gateway policy it also exibits the same problem of stalls and download problems?

6

Hey changeip please come back! I would like to hear your comments.

7

Can you send those results again with the 2 interfaces enabled? Also, it seems to me that once I had 2 separate entries for the default routes and it did not work as expected so we had to put both gateways on the same entry, ie x.x.x.x,y.y.y.y

If all else fails and you want an extra set of eyes to just read over the configs send me an email and I can probably login for a few mins and check things over.

sam at changeip.com

8

Thanks man, you are a true Gentleman :wink:

One entry is the two gateways together as you sugest, but I also have each one as a seperate entry too so I can flick between gateways.

Both work as expected on their own but when load balanced problems occur.

I have done more tests with policy routing too by sending like easynews.com and p2p traffic through the second gateway with all the rest going through the main one. The problem seems to be aparant even then. If for example you have been surfing the net then goto easynews.com which is rule set to use the other gatway there is a long delay of up to 1 minuite and sometimes the browser times out. If you hit refresh it loads or restart the browser it loads after a while!

With loadbalance the problem is worse because sessions are regularly changing gateways and you see the same prolem randomly when this happens.

So I would say that the problem is when the client PC has to change route it occurs. Could it be a new Windows problem? Or is it my config?

I would be very greatfull if you did take a look at my config but I don’t really know what could be wrong?

Is there a problem with 2.8 OS which has been fixed?

It would be good to know from MT if anything like this has been fixed in 2.9

Thanks again I will drop you a mail.

9

I’m using ECMP in 2.9 according to this post about load balancing with a transparent proxy:

http://forum.mikrotik.com/t/how-to-use-transparent-web-proxy-with-2-internet-connections/2257/1

and although it seems to work quite well from a speed perspective downloads are breaking and this is really annoying my customers. I would like some advice or help on ECMP too.