Network traffic distribution


#1

Hi,

We have got an 8-node Aerospike cluster, with around 200K read by seconds. We have a big network issue : we reach the brandwith limit on the network card on one node. Others nodes are ok. Any idea about how to have a better traffic distribution ?

We use lot of UDF. Can it be related ? We have only one namespace, with replication factor of 2, and no disk persistence.

Regards,

Vertrand


#2

Hi

We have got a second cluster, identical to the first one : same issue. One node with around 800 Mb/s TX. One node with around 300 Mb/s TX. Others around 60 Mb/s TX.

Regards,

Bertrand


#3

This is typically observed in scenarios with a hot key.

You may be able to improve the situation by using the feature discussed here: