According to AWS placement group is required to be able to utilize 10gbps network performance. Given that instances are required to be placed in same availability zone for placement group, this sounds risky. For ex, unavailability of one availability zone would bring down the infrastructure.
Considering the risk, why would some use placement group?
Placements groups, as the documentation states, is for those applications were it needs the fastest possible network speeds between your instances, and because of that the instances can't span availability zones.
However, just because you have some of you machines in a placement group doesn't mean you can't also have a duplicate set of machines in another placement group in a separate AZ or even a separate region for your DR / redundancy needs.
A placement group is a physically close group of servers, that are connected via 10 GBE networks (likely the Intel X540-T2)
That group of servers will all be interconnected with 10GBE networks. The next placement group along will also be fully interconnected with 10GBE. However the connection between these groups would need to handle all incoming and outgoing traffic between those groups. If at that point you allowed 10GBE traffic to cross across placement groups you would very quickly run out of bandwidth on the intragroup connections.
For this reason 10GBE is only allowed within a placement group, if availability is a concern, then you would need to deploy into multiple placement groups and have some manner of load balancing between them.
the idea of the placement group is to link instances "locally". so you're right but also the purpose of it isn't to resolve the risk of the single AZ rather to make local servers talk more efficiently to each other.
it may be possible to mix up the env with ELBs and Auto Scaling Groups using placement group parameters to give you the HA over multiple AZs you're after.
Asked in February 2016Viewed 2,541 timesVoted 7Answered 3 times