The two things that make my day! The work on DragonFly-current has led to some significant speed improvements. So good, that Samuel Greear’s post on OSNews.org links to graphed results from him and from Francois Tigeot (multi-page PDF) showing the results from pgbench.
The results show a jump in multi-core/processor numbers that vastly exceeds DragonFly 2.10’s performance, and is comparable to FreeBSD 9/10. Here’s some of what did it.
But wasn’t the original promise of the dragonfly approach that SMP performance would exceed that of the approach of freebsd?
The idea was to approach it in a very different way. FreeBSD 5 brought in a complex locking system to handle SMP, which was problematic for quite a while. DragonFly’s approach has been to shift to simplified lockless or token designs. The advantage of that simpler method is that now that all the parts are in place, it’s much easier to implement this SMP improvement, reliably.
These numbers are comparable to FreeBSD, and while they don’t exceed FreeBSD (and Linux posts the absolute best), it’s a vast improvement compared to what DragonFly 2.10 was doing, and all that improvement was accomplished in perhaps a week or two of work. That’s the ‘easier’ part coming in.