sys_pipe has been modified to avoid contention on DragonFly, which means better performance as tasks get handed between processors. See the commit message for details.
5 Replies to “sys_pipe reoptimized”
Comments are closed.
A running description of activity related to DragonFly BSD.
From the commit message above:
“Performance boosted by around 35%”
With all of these perf improvements I continually read (eg 35% here, 5% just yesterday, etc) – seems like Dfly would be like 200% faster than what it was just a year ago.
Yet I can’t believe that’s true.
What am I missing?
Maybe someone more skilled than me can update the OS benchmark vs FreeBSD etc
This is about performance improvement to a single aspect of the system; so a 35% speedup in one area won’t make the entire system 35% faster.
Still, together and over time they do add up.
If you read the commit messages, a lot of these changes look to be around reducing/eliminating IPI messages (Inter Processor Interrupt), so the effects may be more noticeable on systems with a lot of CPUs (I’m not sure how much it affects multicore and HT processors). Specific types of workload may benefit more than others (basically same thing as Puffy is saying).
Thanks everyone.
Off topic, what are some examples of companies using Dfly?
FreeBSD has Netflix & Whatsapp, etc.
What are some go-to companies using Dfly?
Look at what the percentage change is in – for example, you can’t improve network performance past the maximum speed of the medium being used – but you can reduce latency or the amount of CPU time consumed. That’s three different axes of improvement.