Comments on: Network Latency Inside And Across Amazon EC2 Availability Zones

By: Wes Winham

Wes Winham — Wed, 10 Nov 2010 16:14:10 +0000

I just repeated very similar tests: http://devblog.policystat.com/network-latency-between-ec2-instances-public

It looks like my latency times were fully double what you saw back in May 2009 within the us-east-1b AZ. Has the network really degraded that much since then? What size instances were using and on what OS?

By: Fadi

Fadi — Mon, 23 Aug 2010 21:02:22 +0000

Great data Oren… have you repeated this more recently?
Do you have more recent data?

By: bryanl

bryanl — Tue, 26 May 2009 13:09:36 +0000

Also, ICMP isn’t a good determination of latency due to the possibility of being classified as a low prioritization by the connection routers. I think a better test would be to fire up more instances in different zones and push traffic between them and counting the bandwidth that way.

By: Overview of Amazon EC2 Network Latency | Cirrus Nine

Overview of Amazon EC2 Network Latency | Cirrus Nine — Tue, 26 May 2009 12:52:21 +0000

[...] Good Data [...]

By: Eric Hammond

Eric Hammond — Sun, 24 May 2009 20:10:09 +0000

A single availability zone is like a big data center. The latency between two arbitrary instances within an availability zone could vary depending on how close they are to each other within that data center in the network topology. Another good test would be to fire up 20 instances inside a single availability zone and measure the variability of latency between them.

Also, the names of availability zones are assigned randomly within each account. Your account’s “us-east-1a” may be my account’s “us-east-1c”.