How does round robin gateway load balance mechanism work ?

[deajan]

Hi,

I have that question regarding gateway load balancing strategy.

Let's assume I have two WAN links:
WAN1 = 100Mb/10Mb
WAN2 = 10Mb/1Mb
Both WAN are in a gateway group as Tier 1. Sticky connections is enabled.

Now let's take 20 LAN computers that upload simultaneously some stuff to different destinations.

1/ Does the round robin load balancing let half of the LAN computers pass through WAN1 and the other half through WAN2 ? Meaning that some uploads will be quick, other painfully slow ?
2/ Is there any packet loss detection (when RED or ECN is configured) involved that will make load balancing route more connections to WAN1 instead of a saturated WAN2 link ?

The final question is if the round robin load balance scheduler is affected only by packets to process or also by WAN quality ?

Regards,
Orsiris de Jong.

[jimp]

It is round-robin and connection-based. It knows nothing of bandwidth or usage. You can control the balance with weighting, however.

With sticky, all connections from a specific user are tied to a specific gateway. So in your example with 20 users, if the gateways were wighted 1:1 then you'd likely end up with 10 people using one WAN and 10 people using the other.

If you weight the 100M WAN as 10 and the 10M WAN as 1, you'll get a 10:1 usage ratio (more or less) so you'd end up with ~18 people on one WAN and ~2 on the other. Give or take, might end up 19/1 depending on how the connections come in and the order of the gateways.

Without sticky then you can't predict which connections will end up on which WAN, it would redirect them based on the weight ratio alone.

[deajan]

Thanks alot for your explanation, makes things way more clear for me :)