"Marcin M. Jessa" <lists%yazzy.org@localhost> writes: > On 03/17/2010 01:19 AM, Greg Troxel wrote: >> "Marcin M. Jessa" <lists%yazzy.org@localhost> writes: >> >> >>> I have an AMD64 server running NetBSD 5.0.2 with two 500Gb IDE drives. >>> The drives are raided with RaidFrame and Raid 1. The FFS file systems >>> are mounted with log option. >>> Copying large files i noticed I get about 6-7 mbit/s when running Xen >>> and about 10-11 when running GENERIC kernel and scping to the same >>> file system. >>> I had to disable ACPI on this server as well. >>> Any idea why there is this dicrease in performance ? >>> >> >> My experience was >> >> RAID1 (raidframe) for dom0 fs >> UFS2 fs for backing files >> files created with dd (zero-filled, non-sparse) >> >> I remember dimly >> >> 70 MB/s read from raw raid device >> ~10% less read from file on dom0 >> ~10% less than that read from 'raw device' on domU (rxbd0d) >> >> so your 6-7 vs 10-11 seems like too much. try the above 3 tests and >> post the real numbers. >> > Disabling ACPI and APIC in kernel improved transfer speed over ftp and > scp with 70-80%. > Why is this happening ? You didn't post your results; people can't help without the facts. I would make all three measurements in both cases (stock kernel, and -ACPI/etc.).
Attachment:
pgpl9mWb7sDkw.pgp
Description: PGP signature