What is the primary concern of the paper?

Their primary concern is the number of global accesses of the algorithms, which is the dominant performance impediment on a large multiprocessor.

What is the LH lock for a large multiprocessor?

If code requirements are tight and the authors do not wish the overhead of procedure calls, the LH lock provides good performance for smaller architectures and requires only an atomic swap.

What is the common problem with these locks?

The common problem with these locks, such as the Test&(Test&Set) lock [15], is that they required O(N2) network transactions to service N processors attempting to enter a critical section simultaneously.

What is the advantage of the MCS lock?

Anderson's lock is clearly not competitive any more, requiring O(LP ) memory, a fetch-and-increment primitive, and more global operations than the other locks.

What is the rst acquire/release of the lock?

After the rst acquire/release of the lock, the ag previously owned by the lock is now replicated locally, and the ag owned by the lock resides locally.

How can the authors describe the cost of each critical path for each lock?

By hand-assembling the algorithms into a generic RISC-like assembler, and counting the number and types of operations each execution path requires, the authors can describe the cost of each critical path for every lock.

Why do the authors want to keep the core advantage of the LH lock?

the authors wish to keep the core advantage of the LH lock, namely, that only a single global write is needed to notify that the lock is free.

What is the problem with the MCS lock?

This is in fact one of the problems with the MCS lock, where an \\optimization" is to read the local pointer that, if a queue has formed, will no longer be local.

What is the critical path to improve the LH lock?

The critical paths are:tho = rl + wg + rg + bc + bc=2 (10)topt = 5rl + 4wl + sl + bc (11)tpes = 4rl + 3wl + rg +wg + sg + bc (12)To improve the LH lock the authors need to understand why it requires an additional two global accesses in the pessimistic case.

What is the hand-o cost of the queuing algorithm?

When there is contention for the lock, the hand-o cost will be the part of the sequential execution that will be spent inside the lock algorithm.

Open AccessProceedings Article10.1109/IPPS.1994.288305

Queue locks on cache coherent multiprocessors

Peter S. Magnusson, +2 more

- 01 Apr 1994

- pp 165-171

226

TL;DR: A method to characterize the performance of proposed queue lock algorithms, and applies it to previously published algorithms conclude that the M lock is the best overall queue lock for the class of architectures studied.

Abstract: Large-scale shared-memory multiprocessors typically have long latencies for remote data accesses. A key issue for execution performance of many common applications is the synchronization cost. The communication scalability of synchronization has been improved by the introduction of queue-based spin-locks instead of Test&(Test&Set). For architectures with long access latencies for global data, attention should also be paid to the number of global accesses that are involved in synchronization. We present a method to characterize the performance of proposed queue lock algorithms, and apply it to previously published algorithms. We also present two new queue locks, the LH lock and the M lock. We compare the locks in terms of performance, memory requirements, code size and required hardware support. The LH lock is the simplest of all the locks, yet requires only an atomic swap operation. The M lock is superior in terms of global accesses needed to perform synchronization and still competitive in all other criteria. We conclude that the M lock is the best overall queue lock for the class of architectures studied. >

AI Agents for this Paper

Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps

Most frequently asked questions

1. What is the way to lock a state?

Since elements in the lock array are used sequentially regardless of whether there is a queue or not, the lock must have been taken Pmax times by the local processor for the entire state to be in the cache.

2. What is the time taken for each segment of code?

The time taken for each segment of code is divided up into generic operations: read, write, swap, execute simple instruction, and branches.

3. What is the advantage of the LH lock?

The GT lock also requires O(LP ) memory, and it's advantage of requiring only an atomic swap primitive is equaled by the LH lock.

4. What is the lock for a class of architectures?

The authors conclude that for a class of architectures, large cache-coherent shared-memory multiprocessors, the M lock is the overall preferred choice.

Table 1: Summary of queue lock performance for critical paths

Figure 4: Cache behaviour of the simple LH lock

Figure 3: Enqueueing on the simple LH-lock

Citations

•Book

The Art of Multiprocessor Programming : 並行プログラミングの原理から実践まで

Maurice Herlihy, +2 more

- 01 Jan 2009

TL;DR: This talk will survey the area ofTransactional memory, a computational model in which threads synchronize by optimistic, lock-free transactions, with a focus on open research problems.

...read moreread less

331

•Proceedings Article•10.5555/296806.296814

Tornado: maximizing locality and concurrency in a shared memory multiprocessor operating system

Ben Gamsa, +3 more

- 22 Feb 1999

TL;DR: The design and implementation of Tornado is described, a new operating system designed from the ground up specifically for today's shared memory multiprocessors, which has far better performance characteristics, particularly for multithreaded applications, than existing commercial operating systems.

...read moreread less

241

•Proceedings Article•10.1145/2723372.2746480

FOEDUS: OLTP Engine for a Thousand Cores and NVRAM

Hideaki Kimura

- 27 May 2015

TL;DR: This work develops FOEDUS, an open-source, from-scratch database engine whose architecture is drastically different from traditional databases, which extends in-memory database technologies to further scale up and also allows transactions to efficiently manipulate data pages in both DRAM and NVRAM.

...read moreread less

196

Journal Article•10.1007/S00446-003-0088-6

Shared-memory mutual exclusion: major research trends since 1986

James H. Anderson, +2 more

- 01 Sep 2003

- Distributed Computing

TL;DR: This paper surveys major research trends since 1986 in work on shared-memory mutual exclusion with a focus on algorithms for mutual exclusion.

...read moreread less

166

Proceedings Article•10.1145/2145816.2145849

Revisiting the combining synchronization technique

Panagiota Fatourou, +1 more

- 25 Feb 2012

TL;DR: This paper revisits the combining technique with the goal to discover where its real performance power resides and whether or how ensuring some desired properties would impact performance, and presents two new implementations of this technique which outperform by far all previous state-of-the-art combining-based and fine-grain synchronization algorithms.

...read moreread less

155

...

Expand

References

•Journal Article•10.1145/359545.359563

Time, clocks, and the ordering of events in a distributed system

Leslie Lamport

- 01 Jul 1978

- Communications of The ACM

TL;DR: In this article, the concept of one event happening before another in a distributed system is examined, and a distributed algorithm is given for synchronizing a system of logical clocks which can be used to totally order the events.

...read moreread less

9.4K

•Book Chapter•10.1145/3335772.3335934

Time, clocks, and the ordering of events in a distributed system

Leslie Lamport

- 04 Oct 2019

- Concurrency and Computation: Practice an...

TL;DR: In this paper, the concept of one event happening before another in a distributed system is examined, and a distributed algorithm is given for synchronizing a system of logical clocks which can be used to totally order the events.

...read moreread less

8.4K

Journal Article•10.1109/TC.1979.1675439

How to Make a Multiprocessor Computer That Correctly Executes Multiprocess Programs

Lamport

- 01 Sep 1979

- IEEE Transactions on Computers

TL;DR: Many large sequential computers execute operations in a different order than is specified by the program, and a correct execution by each processor does not guarantee the correct execution of the entire program.

...read moreread less

2.6K

•Journal Article•10.1145/103727.103729

Algorithms for scalable synchronization on shared-memory multiprocessors

John Mellor-Crummey, +1 more

- 01 Feb 1991

- ACM Transactions on Computer Systems

TL;DR: The principal conclusion is that contention due to synchronization need not be a problemin large-scale shared-memory multiprocessors, and the existence of scalable algorithms greatly weakens the case for costly special-purpose hardware support for synchronization, and provides protection against so-called “dance hall” architectures.

...read moreread less

1.2K

Journal Article•10.1109/71.80120

The performance of spin lock alternatives for shared-money multiprocessors

Thomas Anderson

- 01 Jan 1990

- IEEE Transactions on Parallel and Distri...

TL;DR: The author examines the questions of whether there are efficient algorithms for software spin-waiting given hardware support for atomic instructions, or whether more complex kinds of hardware support are needed for performance.

...read moreread less

726

Queue locks on cache coherent multiprocessors

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What is the way to lock a state?

2. What is the time taken for each segment of code?

3. What is the advantage of the LH lock?

4. What is the lock for a class of architectures?

5. What is the primary concern of the paper?

6. What is the LH lock for a large multiprocessor?

7. What is the common problem with these locks?

8. What is the advantage of the MCS lock?

9. What is the rst acquire/release of the lock?

10. How can the authors describe the cost of each critical path for each lock?

11. Why do the authors want to keep the core advantage of the LH lock?

12. What is the problem with the MCS lock?

13. What is the critical path to improve the LH lock?

14. What is the hand-o cost of the queuing algorithm?

Figures

Citations

The Art of Multiprocessor Programming : 並行プログラミングの原理から実践まで

Tornado: maximizing locality and concurrency in a shared memory multiprocessor operating system

FOEDUS: OLTP Engine for a Thousand Cores and NVRAM

Shared-memory mutual exclusion: major research trends since 1986

Revisiting the combining synchronization technique

References

Time, clocks, and the ordering of events in a distributed system

Time, clocks, and the ordering of events in a distributed system

How to Make a Multiprocessor Computer That Correctly Executes Multiprocess Programs

Algorithms for scalable synchronization on shared-memory multiprocessors

The performance of spin lock alternatives for shared-money multiprocessors

Related Papers (5)

Algorithms for scalable synchronization on shared-memory multiprocessors

The performance of spin lock alternatives for shared-money multiprocessors

Wait-free synchronization

Synchronization algorithms for shared-memory multiprocessors

Flat combining and the synchronization-parallelism tradeoff