By using AWS re:Post, you agree to the Terms of Use

AWS RDS Aurora latency baseline

0

Hi, I'm looking to some informations about common latency between EC2 instances and AWS RDS Aurora cluster.

For reference, I have multiple instances running on the same VPC than a cluster and from what we see (tcptraceroute on both writer and reader endpoints) cluster latency is about ~100/200ms if on the same AZ (with some spike at 450ms ... is this normal ?). Results can raised to +800ms if not on the same AZ.

It seems pretty slow to me. What can explain this ? do you have some latency baseline or maybe best practices to share to limit this results.

Regards

  • "multiple instances running on the same VPC than a cluster" - sorry, but I'm a little confused by your description, are your Instances running in the same VPC as the RDS cluster? Or are you routing DB traffic between two VPCs?

1 Answer
0

Hello, In the documentation it says that the latency between different AZ is 1 to 2 milliseconds

https://docs.aws.amazon.com/sap/latest/general/arch-guide-architecture-guidelines-and-decisions.html (See Latency across Availability Zones)

There's a caveat. If you're only using the low-end instances like t2, you will likely have a lot of jitter and can sometimes encounter more latency.

If the instances are ENA (or SR-IOV) compatible, and using it, you'll get a big reduction in actual jitter and can realistically expect very low latency so long as you stay within the same VPC and region.

RDS specifically seems to be well optimized for networking, in the few years I've been using it (us-west-2) I've never had a single problem related to latency with the RDS instances themselves, but this is a N=1 data point.

Also, If you're using many instances connecting to one postgres server, make sure you're using pgbouncer or a similar connection pooler or you'll hit connection issues way before even a few ms of latency will cause problems.

answered 10 months ago
  • Hello, To answer you, we're using t3 instances (ENA enabled) and proxysql.

    So what's acceptable speaking of WriteLatency and ReadLatency ? In our case getting x4 latency just by fetching data from another AZ is questionable (but maybe it's just that Aurora is not the good fit for us)

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions