TéléGC: Remote Garbage Collection using Memory Disaggregation
Team work: Adam Chader presented "TéléGC: Remote Garbage Collection using Memory Disaggregation" at 3a412 the 1/7/2022 at 10h00.
Abstract
Many data analytics frameworks (Hadoop, Spark, Neo4j) are written in managed languages (often Java). A key feature of these languages is *Garbage Collection*, which automatically frees unused memory, preventing leaks.
Datasets are getting increasingly larger, and thus so is the memory footprint of the above-mentioned frameworks. Unfortunately, garbage collection performance scales poorly with memory size, which has an impact on the whole execution of the data analytic job.
We present téléGC, a remote garbage collector, with the property of being unobtrusive, and perfectly concurrent with the execution of the application, by taking advantage of memory disaggregation, and its cache coherency properties.