110 likes | 200 Views
Synchronization Methods for Multicore Programming. Brendan Lynch. Introduction. Synchronization techniques l ocks, barriers, non-blocking approaches, atomic primitives Concurrent memory paradigms s hared memory, d istributed memory Safety concerns and guarantees
E N D
Synchronization Methods for Multicore Programming Brendan Lynch
Introduction • Synchronization techniques • locks, barriers, non-blocking approaches, atomic primitives • Concurrent memory paradigms • shared memory, distributed memory • Safety concerns and guarantees • Mutual exclusion, deadlock, livelock, starvation, race conditions, priority inversion
Synchronization in the Cloud • Granularity of synchronization plays a major roll in parallel performance. Modeling the concurrency of a program allows the user to decide on correct granularity. • Cloud computing relies on a classic distributed memory paradigm. • Memory updates in the cloud are very expensive
Synchronization in the Cloud • Local parallel code should always be optimized for concurrency. • Coarse grain locks provide necessary data synchronization • Design emphasis on reliability and not performance
Chubby Locks • Relies on a locking file system as well as paxos algorithm master election. • These combined give users the illusion of sequential programming • System scales well with development • ‘Young’ systems do not plan for consensus protocol • Maintaining a small file system allows for local small read/write. • Commonly used for synchronizing events with high polling.
Zookeeper • Used by Amazon Simple Queue Service • Claims wait free synchronization through operation linearizability • Coordination kernel allows users to program their own primitive operations. This is important for application specific synchronization requirements. • Zookeeper has two basic ordering guarantees • Linearizable writes and FIFO client order
Zookeeper • Linearizable writes: all requests that update the state of ZooKeeper are serializable and respect precedence • FIFO client ordering: all requests from a given client are executed in the order that they were sent by the client. • Locks and other primitives can be easily built in to the system. Blocking primitives will down-grade offered guarantees; except for the two mentioned above.
Zookeeper • Zab uses by default simple majority quorums to decide on a proposal, so Zab and thus ZooKeeper can only work if a majority of servers are correct (i.e., with 2f + 1 server we can tolerate f failures).
write(2) write(1) Read/Write Register Example linearizable write(0) write(2) read(1) write(1) time time (4) Art of Multiprocessor Programming
write(2) write(1) Read/Write Register Example Not linearizable write(0) read(1) write(2) read(2) write(1) time time (2) Art of Multiprocessor Programming
References • Mike Burrows, The Chubby lock service for loosely-coupled distributed systems, Proceedings of the 7th symposium on Operating systems design and implementation, November 06-08, 2006, Seattle, Washington • Maurice Herlihy and NirShavit. The Art of Multiprocessor Programming. Morgan Kaufmann, March 2008. • P. Hunt, M. Konar, F. P. Junqueira, and B. Reed, "ZooKeeper: wait-free coordination for internet-scale systems," in USENIXATC'10: Proceedings of the 2010 USENIX conference on USENIX annual technical conference. Berkeley, CA, USA: USENIX Association, 2010, pp. 11-11.