17. Synchronization Without Locks
Вставка
- Опубліковано 22 вер 2019
- MIT 6.172 Performance Engineering of Software Systems, Fall 2018
Instructor: Charles Leiserson
View the complete course: ocw.mit.edu/6-172F18
UA-cam Playlist: • MIT 6.172 Performance ...
Prof. Leiserson discusses non-lock synchronization using memory models, Peterson's algorithm, instruction reordering, and hardware reordering. He discusses the compare-and-swap operation; lock-free stacks, pushes, and pops; and the ABA problem.
License: Creative Commons BY-NC-SA
More information at ocw.mit.edu/terms
More courses at ocw.mit.edu
*My takeaways:*
1. Memory model: sequential consistency 0:49
2. Mutual exclusion without locks 12:28
3. Relaxed memory consistency 34:16: never synchronize program through memory
4. Compare and swap 1:06:54
5. Lock free algorithm 1:18:15
Nice and clear. No pauses in between topics and slides. Excellent flow of information thats easily interpreted! Well done Prof!
Very good lecture to get the idea of Memory models and an intro to Lock-free algorithms.
This is a really good course, up there with 6.021 with Erik Demaine!
53:30 In TSO, I think in rule (4), a LOAD may be reordered with a STORE to the the same location, too, if they happen on different cores of a multi-core CPU (so the write buffers are different and the order-preserving load bypass does not trigger). Otherwise, the suggested ordering on slide 29 would be impossible. Took me a while to realize this nuance. Another source I found seems to confirm it, though I'm still not 100% sure this explanation is fully correct.
Nice 👍 database explain popular thank you this position on summer collection
53:55 Oops.. :D I do think he actually used the convention . The student was right to point there was a mistake in the diagram where Load3 is Before Store3, and Load4 is before Store4 thus violating rule 4.
No, he's just numbering the stores and loads. The load3 and store3 just happen to be the third of their kind, and are in this example to be thought of as a load and store to different memory locations.
How to storage working memory handle per files containers and how many for you containers also hear how to for handling
ＲＥＳＰＥＣＴ to the processors order.
Great example/proof performance=currency.
To get this course listen you had to pay $$$ to be sitting in a class in Fall 2018 instead of getting it for free a year later :))
You get much more than just learning materials being in a class. Professors, TAs, peers, labs, exams. All are extremely important for effective learning. It takes a lot of determination to achieve the same level of quality on your own. If you have a chance to get into a good school, don't hesitate it. It's definitely worth it.
What I'm not feeling great about is why is there a linear order? Isn't it assumed that the separate processor instructions are run CONCURRENTLY? Isn't that what the word means!
Execution order 🚚 changes.
1:12:58 Why is the race condition on the variable result??? The variable result is a local variable! Every thread has a separate instance of result! OTOH, myArray has external linkage, and that's the only variable in the example that's possibly shared across threads.
In Cilk, result is actually shared. It's not like pthreads.
@@guidomartinez5099 Oh, that's just unusual.
Need to know what was said at @40.00
I think it was like this:
Teacher: When is it safe to do this reordering?
Student: If A equals B, but if you already have the value 1 in A/B.
Just to be clear, that's why the teacher said, "But more generally when is it safe", because this is very specific to the value of
A, and I think that the only part that would have been able to figure that out is the compiler, if the data is not dynamic.
A and b one group
Hack,try hack, dictionary checking, story buffer
The C++11 standard, not the C11 standard...
In both
en.m.wikipedia.org/wiki/C11_(C_standard_revision)
(pf,coie cookies po files)
Tomorrow meeting "good night "professor thanks for explaining.
Laughing to pariyon class explaining master. 🗃️