Traffic flow in Load Sharing Mode

[Shailendra]

Hi All,

Please help me to understand the packet flow in Unicast mode load sharing mechanism. 

For example in Unicast mode packet is sent  to Pivot and then it either process packet by it self or send it to other member, so my question is - 

(Consider the flow of traffic from LAN to Internet)

1. Which connection link Pivot will use to sent traffic to another member (Sync link or LAN interface link)

2.  when return traffic comes back then will this traffic be sent to Pivot first (If not then on what basis return traffic will reach to non pivot member)

3. Is it that all incoming or outgoing traffic will first processed by Pivot and non Pivot will always get new request from Pivot.

Thanks 

Regards

Shailendra

[_Val_]

Traffic is always received by pivot member, all incoming packets, both from client and server side, and then it may be forwarded to other cluster members, according to decision function. After packet is being filtered by the actual cluster member, it is sent to the next hop directly and not through pivot.

Traffic is always forwarded to the next cluster member on the same interface it is received and not through sync. 

I would strongly consider to avoid this mode, unless absolutely required (still looking forward to seeing any justification here 🙂 )

[Kaspars_Zibarts]

It was 2007 when we used it in cluster of 3 Nokia IP530 boxes. Actually worked like a charm, better then broadcast method for some reason 🙂 must have been R55

[_Val_]

In early 200x, sometimes adjacent switches did not go well with multicast modes, so unicast LS was the only alternative. By no means it should not be the case today, and also, classic ClusterXL LS mode should not be considered a proper performance optimization approach.

[the_rock]

You got 2 very valid responses. I believe all this should be explained in clustering guides as well.

[_Val_]

It is explained there, I checked yesterday with R81 guide 🙂

[PhoneBoy]

If you’re considering ClusterXL Load Sharing, I would strongly suggest bigger gateways in HA or using Maestro.