On Wed, Apr 06, 2011 at 04:39:04PM +0100, Mindaugas Rasiukevicius wrote: > > No, I do not see the need to collect statistics about every allocation, > e.g. load of temporary buffer, string, whatever allocations/frees.
Besides which, DTrace can *do* this. And, contrary to what I've seen in this thread, there is no great mystery about how to implement DTrace support for more architectures. We did the i386 implementation with no more documentation than what Sun supplies and the public discussions of how it had been ported to FreeBSD. On the other hand, someone highly skilled will need to do the work -- someone familiar both with the architecture for which support is being implemented and with our kernel as a whole, and willing to read and understand the DTrace code with the help of Sun's documentation. We (Coyote Point) can't do that right now, because our people with the right skills are tied up elsewhere. However, Darran Hunt, who did the original work for us, would likely be more than glad to help anyone seriously interested in implementing it on other architectures. I submit that DTrace can do what the majority of our existing ad-hoc debugging and diagnostic frameworks can do, and much more. Developers with the skills to devise and implement new such code might find that for the same amount of work, a greater payback came from working on DTrace instead. Thor
