Peter Rudenko

Do you have yarn logs for executors?

The issue is that disni reuses wc object. In your code you need to do: ```java lastEvent.set(wc.clone()); ``` After changing this got correct result SEND -> RECV -> SEND ->...

@yaelna Can you please run this function: ```java public void dispatchCqEvent(IbvWC wc) throws IOException { System.out.println("Server got event " + wc.getWr_id() + " : " + IbvWC.IbvWcOpcode.valueOf(wc.getOpcode()) + "Old "...

It includes libraries into jar: https://github.com/zrlio/disni/pull/17/files#diff-6c01060f17a19807e7ef9d8221b3e43aR51 In assembly jar you'll see /lib folder with binaries. Need to think maybe to do cross compilation for different platforms + something like this:...

In assembly jar there's only libdisni binary, all other libraries (spdk, dpdk, etc.) should be managed by LD_LIBRARY_PATH. The issue if use disni as dependency library need to have 1:1...

No need to extract to some folder. It was just an example. Totally possible to read from jar itself. Here's for example netty (not hello world app): https://github.com/netty/netty/blob/4.1/transport-native-epoll/pom.xml#L171 Apache Ignite:...

Would be great to measure performance loss in pyspark vs scala for mllib models implemented in scala.

Any plans to test performance new ml API (Pipeline, Crossvalidation, GridSearch, etc.)?

The current version of SparkUCX utilizes RDMA to accelerate shuffle block transfer. We're working on a version, that would have additional protocols, so it should work with TCP transport.

We expect to have it in a few months.