Product · Live Technology · 2018
Live Platform
Search and caching architecture for a Fortune 500 advertising platform, cutting average request time by over 68 percent.
How it works
Caching on both sides of the request, with the search index between them. The bottleneck was the query path, so that is where the work went.
-
01 · The problem
Client-facing data requests were taking 450ms on average. At the platform's request volume that was the difference between a campaign that scaled and one that did not.
-
02 · What I decided
Profile before optimising. The bottleneck was the query path, not the transport, so the work went into proprietary search algorithms and a distributed cache on both sides rather than into more infrastructure.
-
03 · The outcome
Average request time down from roughly 450ms to 125ms, a reduction of over 68 percent, holding under heavy load for clients including Bank of America, Century 21, and Northwestern Mutual.
What a surveyor, or a hiring panel, should hear.
Constraint
Average request time of 450ms at the platform's volume was the difference between a campaign that scaled and one that did not, for clients who would not tolerate either.
Decision
Profile before optimising. The bottleneck was the query path, not the transport, so the work went into search algorithms and a two-sided cache rather than into more infrastructure, which would have cost more and fixed less.
Ownership
Down to 125ms, a reduction of over 68 percent, holding under load for Bank of America, Century 21, and Northwestern Mutual. Promoted to senior within three months on that work.
- Latency
- 450ms to 125ms
- Cut
- 68%
- Platforms
- Web