By: Michael S (already5chosen.delete@this.yahoo.com), July 26, 2007 4:22 am
Room: Moderated Discussions
Foo_ (foo@nomail.com) on 7/26/07 wrote:
---------------------------
>Paul (no@thanks.com) on 7/25/07 wrote:
>---------------------------
>>Given that stuff that was regarded as "Enterprise" a few years ago such as decent
>>SMP scaling (and SMP support at all), >1GB memory support on x86 and even NUMA.
>>Is now "Desktop" I would say the position of having a keeping a single kernel for
>>everything (with a few configurable options) rather than trying to tweak for a particular
>>segment as Con was trying to do has been entirely vindicated.
>
>Uh, the specificity of desktop vs. "enterprise" is not in the hardware but in the
>usage model. You can use the same hardware as both desktop and "enterprise", use cases will still be varying different.
>(see Linus' response in this thread, and also Con's interview as linked to in the first post)
>
According to my understanding, Paul meant to say that with good scalable SMP support when you have 4 cores (often even 2 cores) the CPU scheduling algorithm have negligible effect on desktop responsiveness.
Same with memory - when an average desktop has >1GB in place kernel developers would rather not spend their precious time fine-tuning swapping vs disk cache.
It seems to me that when mighty 4-core, 2GB desktop feels slow in most cases you want to blame either slow disk or crappy application design rather than "kernel overtuned for enterprise".
---------------------------
>Paul (no@thanks.com) on 7/25/07 wrote:
>---------------------------
>>Given that stuff that was regarded as "Enterprise" a few years ago such as decent
>>SMP scaling (and SMP support at all), >1GB memory support on x86 and even NUMA.
>>Is now "Desktop" I would say the position of having a keeping a single kernel for
>>everything (with a few configurable options) rather than trying to tweak for a particular
>>segment as Con was trying to do has been entirely vindicated.
>
>Uh, the specificity of desktop vs. "enterprise" is not in the hardware but in the
>usage model. You can use the same hardware as both desktop and "enterprise", use cases will still be varying different.
>(see Linus' response in this thread, and also Con's interview as linked to in the first post)
>
According to my understanding, Paul meant to say that with good scalable SMP support when you have 4 cores (often even 2 cores) the CPU scheduling algorithm have negligible effect on desktop responsiveness.
Same with memory - when an average desktop has >1GB in place kernel developers would rather not spend their precious time fine-tuning swapping vs disk cache.
It seems to me that when mighty 4-core, 2GB desktop feels slow in most cases you want to blame either slow disk or crappy application design rather than "kernel overtuned for enterprise".
Topic | Posted By | Date |
---|---|---|
Linux on Desktops | Shankhadeep | 2007/07/25 10:46 AM |
Linux on Desktops | Paul | 2007/07/25 11:42 AM |
Linux on Desktops | Shankhadeep | 2007/07/25 02:15 PM |
Linux on Desktops | Max | 2007/07/25 02:43 PM |
Linux on Desktops | JasonB | 2007/07/26 11:38 PM |
Linux on Desktops | Foo_ | 2007/07/26 02:53 AM |
Linux on Desktops | Michael S | 2007/07/26 04:22 AM |
Linux on Desktops | Paul | 2007/07/26 01:56 PM |
Linux on Desktops | Foo_ | 2007/07/26 02:34 PM |
Linux on Desktops | Jukka Larja | 2007/07/27 04:42 AM |
Linux on Desktops | Shankhadeep | 2007/07/27 11:52 AM |
Linux on Desktops | Shankhadeep | 2007/07/27 12:00 PM |
Linux on Desktops | _Arthur | 2007/07/27 02:56 PM |
Linux on Desktops | Linus Torvalds | 2007/07/25 12:03 PM |
Linux on Desktops | Shankhadeep | 2007/07/25 01:36 PM |
Linux on Desktops | Arun Ramakrishnan | 2007/07/25 08:11 PM |