Practical Considerations for Non-Blocking Concurrent Objects

An important class of concurrent objects are those that are non-blocking, that is, whose operations are not contained within mutually exclusive critical sections. A non-blocking object can be accessed by many threads at a time, yet update protocols based on atomic Compare-And-Swap operations can be used to guarantee the object's consistency. In this paper we take a practical look at the Compare-And-Swap operation in the context of contemporary bus-based shared memory multiprocessors, although our results generalize to distributed shared memory multiprocessors. We rst describe an operating system-based solution that permits the construction of a non-blocking Compare-And-Swap function on architectures that only support more primitive atomic primitives such as Test-And-Set or Atomic Exchange. We then evaluate several locking strategies that can be used to synthesize a Compare-And-Swap operation, and show that the common techniques for reducing synchronization overhead in the presence of contention are inappropriate when used as the basis for nonblocking synchronization. We then describe a simple synchronization strategy that has good performance because it avoids much of the synchronization overhead that normally occurs when there is contention. This research was sponsored in part by a National Science Foundation Presidential Young Investigator Award, the Digital Equipment Corporation, the Defense Advanced Research Projects Agency, Information Science and Technology O ce, under the title \Research on Parallel Computing", ARPA Order No. 7330, issued by DARPA/CMO under Contract MDA97290-C-0035, and by the Open Software Foundation (OSF). The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the o cial policies, either expressed or implied, of NSF, DEC, DARPA, OSF, or the U.S. government.