Scalability prediction
47 views
Skip to first unread message
Amine Moulay Ramdane
Aug 7, 2015, 4:50:32 PM8/7/15
to Scalable Synchronization Algorithms
Hello...
As you have noticed i have implemented a concurrent Skiplist, the interface of my concurrent Skiplist is very fexible,
and as you have noticed i am using one distributed reader-writer mutex to implement my concurrent algorithm, so i am not using RCU or SMR or VZoom, so i have done a scalability prediction for this algorithm, and i have noticed that on a 32 NUMA nodes computer and 50 millions of read , write and delete requests, with a workload of 0.5% of writes and ~99% of reads, it will give a throughput of 50 millions request per 10 seconds , that's cool.
On a non-NUMA Quadcore it will give 4X scalability.
You can download my new concurrent Skiplist version 1.02 from:
https://sites.google.com/site/aminer68/concurrent-skiplist
Thank you,
Amine Moulay Ramdane.
As you have noticed i have implemented a concurrent Skiplist, the interface of my concurrent Skiplist is very fexible,
and as you have noticed i am using one distributed reader-writer mutex to implement my concurrent algorithm, so i am not using RCU or SMR or VZoom, so i have done a scalability prediction for this algorithm, and i have noticed that on a 32 NUMA nodes computer and 50 millions of read , write and delete requests, with a workload of 0.5% of writes and ~99% of reads, it will give a throughput of 50 millions request per 10 seconds , that's cool.
On a non-NUMA Quadcore it will give 4X scalability.
You can download my new concurrent Skiplist version 1.02 from:
https://sites.google.com/site/aminer68/concurrent-skiplist
Thank you,
Amine Moulay Ramdane.
Reply all
Reply to author
Forward
0 new messages