> Hi everyone > I'm a final year student in computer science at Universidad de Concepción, > Chile. My graduation project is about enhancing Memcached performance > using wait-free techniques in order to improve multithreading scalability. > I would like to know if someone has tried this before and what do you think > about this approach too. > > Regards.
We've had a lot of different approaches that people have patched over the years. They're not all wait-free, but all try to scale the central locks in different ways. There was an RCU implementation (high read speed, low write speed), Intel's recent engine (which I haven't examined the source of yet), the one from ... the solarsomething network company. That one was similar to what I ended up with, but it increased memory usage. I was experimenting with the current system the other day and got it up past 11 million key fetches/sec, which I think can be made into a safe change. So I'm saying there've been a bunch of attempts to scale the central locks but not sure how many are specifically "wait free". Did you have a particular algorithm in mind? :P