svar, I have done some tests with concurrent load but did not see performance drops. It …

… INSERT benchmark What did you mean by that? In your tests you ran many INSERT .. ON DUPLICATE KEY UPDATE queries. That… to work pretty well (better than InnoDB in the 2nd test) for these upsert queries. So why did you say the…

Thanks Vadim, I did the test on 5.6 and load data infile concurrently results for … key worth distribution, memory or CPU as repeating the same test on uncompressed table scale as expected. Mark did you cover…

…, which is available in MySQL 5.6. So we may test MySQL 5.6 more to see what kind of performance…

RIch, Like I said, unfortunately upsert part did not work in this benchmark, so it was mostly INSERT benchmark. I fixed workload and right now I am running a different test, which uses “x=x+value(x)” a lot. So I am very interested to have it optimized and would like to run it.

Rich, Was this upsert optimization used in Vadim’s tests? If not how do you turn this optimization on?