1

Amazon elasticache cluster looks to be a really good choice for auto-scaling but it seems the load distribution is not equal among the nodes.

Both memory and CPU wise the node 1 seems to take almost twice the resource than other nodes. This often leads to increase in response time at peak hours even when most other nodes are sitting ideally.

I am using r3.2x instances (8cpu, 60GB mem)

Is anyone having the same problem? What can be cause for this behaviour (in elasticache or in my memcache module)? On my side I am using distribution consistent hashing (Ketama algo). Any better alternative ?

Ashish Gaurav
  • 111
  • 3

1 Answers1

0

If you have one single cache-item (or very few items) that is used much more often than any other item, you will see this behavior of uneven load distribution.

You should check your application for this kind of cache-items. Perhaps you can log all memcache-requests to find these items. After you identified such item you have to think about how to split and distribute calls to this item to many items.

I believe there is no single solution for this kind of problem because it depends on the application-logic, the number of reads and writes, etc.

heiko
  • 1,733
  • 1
  • 12
  • 6