M
M
MHNET2021-08-25 17:50:21
Cisco
MHNET, 2021-08-25 17:50:21

Which load balancing to choose?

So, there is the following network:
Many groups of Cisco access switches with aggregated links (LACP) between them (between switches in the group (Po2)) - one of the access switches in each group is connected to the L3 stack of Cisco switches. The stack of switches, in turn, is connected to a router (or rather, to the Fortigate firewall) with many ports with subinterfaces on each, aggregated in Po3 (LACP). Servers are also connected to the L3 stack of switches, the links between which are also planned to be aggregated. Which load balancing method to choose:
1. Between L2 switches
2. Between L2 switch and L3 stack
3. Between L3 stack and servers
4. Between L3 stack and router
I read that between the servers and the L3 stack, in this case, you need to balance using DST-MAC or DST-IP. If so, what about the rest of the devices? And where can I read in detail about the recommended options for load balancing?

Answer the question

In order to leave comments, you need to log in

1 answer(s)
V
Valentine, 2021-08-25
@vvpoloskin

Get your head up. When aggregating ports, the device makes a hash from a certain parameter, and based on it, it scatters packets on certain ports. Here is an aggregated port between two routers. The MAC address hash for all packets will be the same, everything will go to one physical port. A hash by IP address will be better, since as a rule, packets from different IP addresses come to the router.
Further, when used between access and aggregation switches. As a rule, there are identified traffic gravity points (video services). All users are split to one IP address. But they all have different Mac addresses. In this case, balancing on src-Mac will be better.
Previously, the choice of balancing method was a creative process and was chosen empirically. We turned on the algorithm, looked at the MRTG charts a day later, didn’t like it, turned on another one, looked at the charts and meditated on even balancing. It was so because everything is different for everyone: there are BGP balancing, there are nats, there is MPLS, there is routing on aggregation-access ... now it seems that all SUVs have the src-dst-mac-ip algorithm. And only in the MPLS environment does hemorrhoids begin with its flow-label. But that's a completely different story.

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question