Messages in this thread | | | Date | Tue, 7 Sep 1999 01:15:32 -0700 (PDT) | From | Robert Dinse <> | Subject | CPU scaling in Linux 2.2.x |
| |
I've built non-SMP kernels for the two SS-10's here as suggested by David Miller, which running on a single CPU, are working fine but slower. But I thought it was interesting how much difference what the machines are doing made in terms of the benefits of having multiple CPU's.
The web server, with 4 CPU's running 2.2.10, with Andrea's schedualer patch, and the others I mentioned, was running around 15% CPU occupancy, with one CPU only about 25%, so it doesn't benefit a whole bunch from multiple CPU's. The news server on the other hand went from 9% to slightly more than 25% and the latency went up substantially (and NNTP is very sensitive to latency and thus it's ability to keep up with an incoming feed was substantially hurt).
I just thought it was interesting how a machine doing a lot of network I/O barely benefits from the additional CPU's, but one doing a lot of disk I/O benefits greatly. For the web server it strongly suggests I'd be better of getting a single 160mhz module verses quad 100's and that's a lot cheaper than replacing the whole machine. But the news server is another matter.
Has anybody tried 2.3 kernels on Sparc-32? Last time I grabbed a 2.3 kernel it wouldn't compile but I haven't done so recently.
- 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/
| |