The yellow line is a hardware accelerated transactional memory system.The system can figure out which operations conflict and which not, and only necessary synchronisation is added. Hence, we can observe that in the balanced case scenario, the TM system performs better than fine-grained as it does not necessarily take locks at each node, whereas fine-grained locking system does.
MichaelJordan
For hashmaps, the TM system is about as good as fine-grained locking because there is just one lock that needs to be taken (per bucket in the hashmap), regardless of which method you use for synchronization.
rohany
The TM system allows for significantly better performance on the balanced tree case, because the transactional memory system allows for concurrent operations that would ordinarily require multiple lock takes to occur concurrently
pdp
In a balanced tree, when read sets of two transactions are nearly the same but write sets different, TM system will allow for more concurrency and better performance than fine locks.
lfragago
It is noteworthy that for balanced trees when there is just one or two processors, the execution time is larger using fine locks than it is using coarse locks. The reason behind that is because with fine grained locks we have more overhead of locking unlocking, so with fewer processors this overheard overshadows the benefit of fine granularity.
jedi
What is the fraction of conflicting operations for these two charts?
fxffx
This graph shows that fine locks have some overhead and are slower than coarse locks when processors number are small, but the benefit of fine lock increases as the number of processors increases.
The yellow line is a hardware accelerated transactional memory system.The system can figure out which operations conflict and which not, and only necessary synchronisation is added. Hence, we can observe that in the balanced case scenario, the TM system performs better than fine-grained as it does not necessarily take locks at each node, whereas fine-grained locking system does.
For hashmaps, the TM system is about as good as fine-grained locking because there is just one lock that needs to be taken (per bucket in the hashmap), regardless of which method you use for synchronization.
The TM system allows for significantly better performance on the balanced tree case, because the transactional memory system allows for concurrent operations that would ordinarily require multiple lock takes to occur concurrently
In a balanced tree, when read sets of two transactions are nearly the same but write sets different, TM system will allow for more concurrency and better performance than fine locks.
It is noteworthy that for balanced trees when there is just one or two processors, the execution time is larger using fine locks than it is using coarse locks. The reason behind that is because with fine grained locks we have more overhead of locking unlocking, so with fewer processors this overheard overshadows the benefit of fine granularity.
What is the fraction of conflicting operations for these two charts?
This graph shows that fine locks have some overhead and are slower than coarse locks when processors number are small, but the benefit of fine lock increases as the number of processors increases.