Detta är ett uppsatsförslag hämtat från Nationella Exjobb-poolen. Klicka här för att komma tillbaka till samtliga exjobbsförslag.
Optimizing matrix multiplication in large Galois Fields
Degoo is developing a P2P Backup system which has its roots in the Master’s Thesis “CollabBackup: A Peer-to-Peer backup system focused on storage efficiency”. There’s still some work needed before the system is ready for the commercial market. This Master’s Thesis is an important part of that.
In order to maximize the reliability of the system we are using a special instance of erasure codes called network codes. Network codes have very useful properties when restoring the replication-degree in the system but they suffer from quite bad performance. We therefore must make our implementation as fast as possible, in order to improve the user experience. This work boils down to optimizing matrix multiplication in large Galois Fields. We currently have a SSE-optimized version that provides reasonable performance. However, we still that it can be improved upon by doing the following:
1. Implementing a GPU-accelerated version. Matrix multiplication is very easy to run in parallel so a GPU should be able to beat a CPU quite easily.
2. Improving the cache-hit ratio of the CPU-version.
Informationen om uppsatsförslag är hämtad från Nationella Exjobb-poolen.