We have a new type of drmemtrace: core-sharded-on-disk. We'd like to run the invariant checker on such a trace. We'd also like to run it as an additional tool in dynamic scheduling core-sharded mode of a thread-sharded-on-disk trace.
Some existing invariant checks assume every part of a software thread is seen by one checker thread. This breaks down with core-sharded where threads are interleaved. We'd prefer to give up some of these checks and maintain parallel operation than serialize the cores. We should also think about the core's view and insertion of kernel pieces (we already have features to insert context switch snippets and syscall snippets).
We have a new type of drmemtrace: core-sharded-on-disk. We'd like to run the invariant checker on such a trace. We'd also like to run it as an additional tool in dynamic scheduling core-sharded mode of a thread-sharded-on-disk trace.
Some existing invariant checks assume every part of a software thread is seen by one checker thread. This breaks down with core-sharded where threads are interleaved. We'd prefer to give up some of these checks and maintain parallel operation than serialize the cores. We should also think about the core's view and insertion of kernel pieces (we already have features to insert context switch snippets and syscall snippets).