Hyaline: Fast and Transparent Lock-Free Memory Reclamation
share
We present a new lock-free safe memory reclamation algorithm, Hyaline, which is fast, scalable, and transparent to the underlying data structures. Due to very low contention, Hyaline is generally faster than existing approaches, including epoch-based reclamation. Moreover, Hyaline easily handles virtually unbounded number of threads (or any concurrent entities) that can be created and deleted dynamically, while retaining O(1) reclamation cost. Hyaline also does not depend on any OS abstractions or compiler support. Hyaline's full transparency and simple API enable easy integration into unmanaged C/C++ code, while not causing any undue burden on programmers, potentially making the entire process fully automatic through special language constructs. We also extend Hyaline to avoid situations where stalled threads prevent others from reclaiming newly allocated objects, a common problem with epoch-based reclamation. We have implemented and tested Hyaline on the x86(-64), ARM32/64, PPC, and MIPS architectures. The general approach typically requires LL/SC or double-width CAS, while a specialized version also works with single-width CAS. Our evaluation reveals that Hyaline's throughput is very high -- it steadily outperformed epoch-based reclamation by 10 gains in oversubscribed scenarios.
READ FULL TEXT
