Hi, I am working on implementing and comparing performance of an in-memory database engine on Intel's Xeon Phi processor. However, being new to this, was wondering if there is anyone in this community who have worked on this before? I would really appreciate help and if possible contact you directly for further discussion. Some issues I am concerned with is leveraging the cores for the database transactions, etc. along with compilation problems. We have successfully been able to use virtual distributed system for the same but we now want to compare the performance of running it on Xeon Phi processor. Perhaps link to any documentation etc would be helpful.
I don't have a lot of experience in running in-memory DBs on a Xeon Phi. However, I'd expect anything non-floating point to perform quite poorly on the current generation of Phis. All benchmarks that I've thrown at a Xeon Phi that were not properly parallellized&vectorized resulted in atrocious performance compared to a "regular" Xeon. The cores on the Phi are not all independent cores but require delicate interplay to get them to perform optimally - I'd expect that it will be very hard (of not at all impossible) to achieve this for a database-like application.
Of course, this is just my 2 cents worth ;)