Linkerd is a "service mesh" proxy on the JVM, for which memory footprint and tail latency are critical concerns. Thus, we'd love to try AOT-compiling Linkerd to see just how fast and small we can make it.
Our initial experiments (linkerd/linkerd#1935) suggest that this will require some SubstrateVM changes to get Finagle working.
We've built a basic perf-testing harness here that might be useful as a starting point (currently has OpenJDK and OpenJ9): https://github.com/linkerd/linkerd-examples/tree/master/perf-baseline/linkerd1-perf
wmorgan
👍3
>All comments
We have looked into linkerd. The blocker is allocating threads in static initializers. We are working on a solution for delaying some class initialization until runtime to avoid these kind of issues.
cstancu
on 7 Aug 2018
👍3
Was this page helpful?
0 / 5 - 0 ratings
Related issues
ghost
路
3Comments
koduki
路
3Comments
guaporocco
路
3Comments
newk5
路
3Comments
gersonimo
路
3Comments
Most helpful comment
We have looked into linkerd. The blocker is allocating threads in static initializers. We are working on a solution for delaying some class initialization until runtime to avoid these kind of issues.