Messages in this thread | | | Date | Fri, 21 Jan 2000 11:33:33 +0100 | From | Helge Hafting <> | Subject | Re: Interesting analysis of linux kernel threading by IBM |
| |
Davide Libenzi wrote: > > Hi Helge, > > Friday, January 21, 2000 9:03 AM > Helge Hafting <helgehaf@idb.hist.no> wrote : > > Even better: dynamically use the best algorithm depending on the number > > of > > runnable processes. The only problem I see with this is that calling > > the > > scheduler through a pointer (or if-statement) will have a overhead > > of its own - every time. > > You don't need to change the scheduler under normal environments since the > patch > perform equally in these cases giving better results under high load.
I have the impression that you wrote this: >Is my thought that this patch must be included inside the scheduler since it >gives higher performances >with a lot of tasks and a 15 % less with 2 tasks ( at the same amount of >switches / second )
earlier in the thread. I have seen the argument before that such a loss isn't accaptable because most machines only have a few runnable threads. The common case may have thousands of threads, but only a handful ready to run, with the rest blocked on io. (io being network or disk, a networking benchmark doing local communication only isn't realistic, and many databases cannot be held entirely in memory eiter.)
This is why I suggested the dynamic approach, the few who actually get more than a handful of threads ready-to-run *most of the time* may then get better performance without punishing the vast majorities who seldom see a load above 2.
Helge Hafting
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |