Messages in this thread | | | Date | Tue, 27 Jul 1999 17:12:29 -0700 (PDT) | From | Gerald Aigner <> | Subject | Re: (disk/cpu) kernel performance problem |
| |
On Wed, 28 Jul 1999, Stephen C. Tweedie wrote:
> Hi, > > On Mon, 26 Jul 1999 20:37:26 -0700 (PDT), Gerald Aigner > <aigner@stanford.edu> said: > > >> If disk I/O were infinitely slow, CPU usage would be 0%, > >> but that's not a Good Thing, is it.. > >> > > 45% is the amount of time the kernel spends in loading data from disk. > > Umm, you _have_ enabled IDE DMA, haven't you, either by hdparm -d or by > enabling it by default in the kernel? 45% seems rather high unless > you're using polling transfers. >
Yes, DMA is enabled. Please keep in mind that I am reading something like 25-30 MB/second from the disks.
I know that 45% (or in the case of 2.3.11 37%) is high and that's why I ask. I really would like to have these cycles to do my calculations.
I thought that mmap would help since the kernel wouldn't have to do any (?) memory to memory transfers but unfortunately the kernel seems to serialize mmap accesses in different threads of the same process. I suspect that current->mm->mmap_sem is the bad semaphore :-).
Thanks for any help, Gerald
- 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/
| |