Pivotal Knowledge Base

Follow

How to measure GemFire WAN latency

Environment

Product Version
Pivotal GemFire 7.0.2.x - 8.2.0.x

Purpose

In GemFire, there is no publicly supported way to get information on when a key was inserted and propagated into the other clusters in the WAN. This article describes how latency across the GemFire WAN can be recorded.

Procedure

Currently, the recommended way to measure WAN latency is to use a technique used by many traditional, replicated databases, or file stores. Using an extra region, you can put a timestamp into the region at a regular interval on one end of the replication. On the other end, have a timer event check the current timestamps in the region. This technique has been used successfully for both file system replication and database replication. The key should be an ID for the data center, so that all data centers (GemFire clusters) can insert their time stamp into the specified region. Then, at any point in time, data center X can look at the region and see entries for data center X, Y, and Z, as well as the last time they were replicated. The difference to the current time gives the latency.

For example, given the following values:

Current time: 1000
Key            Value
X              1000
Y              950
Z              940

the latency, from X's perspective, between X and Y is 50ms, and between X and Z is 60ms.

Comments

Powered by Zendesk