- 新しい順
- 投票が多い順
- コメントが多い順
InterZone-In and InterZone-Out are cross AZ data transfer costs. We often see this in containerized workloads when each POD is Multi-AZ, when the communication from 1 POD to another can cross AZs, then you can end up with this ping-pong type affect of traffic crossing back and forth between AZs. I suspect this is what you are seeing.
The only way I know of to limit this is to only have the Multi-AZ aspects of the workload at the ingress LB and the data store (DB?) and have each AZ deploy a set of PODs. This requires that the workload activity would be quick enough such that on a retry and timeouts if an AZ were to fail, the client would quickly try again and the next request would go to another AZ. You would still have some cross-AZ traffic at the LB and DB layers, but the POD to POD communications would all be in the same AZ.
Hi, do you have an ECR in each region with replication of container images for ECS? That may be a source of trafic
Actually no, we are using Docker Hub
関連するコンテンツ
- AWS公式更新しました 1年前
- AWS公式更新しました 7ヶ月前
- AWS公式更新しました 2年前
- AWS公式更新しました 3年前
Thanks, Shlomo! We'll try this approach.
Please let me know how it goes. ;-)
Things are working now as expected. Thanks! :)