Hello networkers !
As you know, there are plethora of different TCP variants/algo/optimizations. However, most people tend to leave default implementation as, in most cases, it performs pretty well.
That said, in some cases (Datacenters, high loss environments, very high latency/RTT environments,...), TCP variants could bring, in theory, good improvments. For example, I heard (years ago) that Microsoft got nice results by using ECN in its datacenters. And of course, everybody heard that Google works a lot on things like BBR or QUIC (but Google's network is more like a "research" huge environment than a "real world" use case ;-) ).
So I'm wondering :
- Today, in 2019, did you specifically chose to implement a TCP variant ?
- Why and in which environment ?
- Was it a success ? A failure ?
- Do you have some metrics that demonstrates the effectiveness of your choice ?
Thanks !
No comments:
Post a Comment