(Cross posted from the txtWeb Engineering Blog)

At txtWeb, we were faced with a hard problem : handling peak loads during the IPL season without impacting our end to end response times. The End to End response time chart(E2E) below is an indicator of how our latencies were increasing with load:

As you can observe, it is very clear that as the requests per day increased, the latency also shoots up, which means that our end users are getting back delayed responses. This was an issue because the platform was able to scale up to the huge loads which txtWeb was receiving, but the delay was happening after the response message leaves the platform. This was a very important problem for us to solve, as this effectively meant on all the days we get huge number of requests, our end users would receive delayed responses.

Simply put,

The Problem:

• A single number - 9243342000 was being bombarded by millions of requests  a day.

Our solution:

• Load balance our users and get local numbers.

What we did:

• Got 25 different numbers for cities across India.
• Especially those states which were participating in the IPL.