How much latency will moving Redis session storage to a managed service add?

[EricHG]

I'm trying to decide if we should move our stateful session storage from our on-prem Redis cluster to a managed database service, but I'm worried about the latency impact on user experience. Our current setup is tightly coupled with the app servers, and I'm not sure if the network hop to a cloud provider's managed service will introduce noticeable lag during peak traffic.

[Gregory.S]

We tried a move last quarter from our on prem Redis to a managed service and saw latency creep up at peak. The app servers got one extra hop and p99 latency rose by tens of milliseconds in the worst hours. We ran a small pilot in a single region and tracked response times under load, which convinced me to slow down and plan more changes before a full cutover.

[IsabellaYR]

After we eventually switched we added a direct peering path and kept the most active sessions on a nearby region, and the spikes softened a bit. Still the initial hit for new sessions felt noticeable in heavy traffic but not fatal.

[JohnL]

Maybe the real bottleneck was something else like how we serialize session data or how often we touch the store during a request. In our tests the storage lag did not always line up with user perceived delay, which was confusing and made me wonder if we were chasing the wrong problem.

[Amelia54]

Would a colocated or hybrid approach help reduce risk while we test latency sensitive paths?