This post is for recording some notes from a few OSDI'21 papers that I got fun.
DMon: Efficient Detection and Correction of Data Locality Problems Using Selective Profiling https://www.usenix.org/system/files/osdi21-khan.pdf
DMon is a compiler-based data locality optimization system. The approach is quite similar to that of daphicx or similar things. It injects data access profiling code in the target program, build and run it with some workload, collect the profiling code-generated results, analyze the profile results to figure out what kind of optimization can provide some benefit to the program, injects the optimization code in the program, built it again, and deploy the version to the production.
I set watching at least one OSDI'20 presentation video per day during the long vacation as one of my plans.
A large scale analysis of hundreds of in-memory cache clusters at Twitter https://www.youtube.com/watch?v=OQtMM5vdhlI&feature=emb_title
The authors traced and analyzed the workloads on the Twitter’s in-memory cache systems. To me, below findings were interesting.
There are many write requests. Size of each object is not so big (median 200 bytes), so metadata, which sizes 64 bytes per each object, is a burden Size of each key is not small compared to the size of each value.