RE: [Xen-ia64-devel]PATCH] Remove duplicate check is_running_on_xen

["Xu, Anthony" <anthony.xu@xxxxxxxxx>]

INAKOSHI Hiroya write on 2007年1月16日 12:50:
> Hi,

Hi hiroya,

Thanks for testing this.
This report is professional,

Intel is working on user level hypercall,
But we can't see any improvement from benchmark FileIO.

And we did see ITC cycle of hyercall from Qemu reduces from 1750 ITC cycle 
to 1000 ITC cycle.

We think user level hypercall can improve IDE disk performance.
Do you know any benchmark we can use to test IDE disk performance?

Thanks,
Anthony


> 
> here is a result evaluating the series of Anthony's patches.  The
> throughput of syscall in unixbench has been lifted more than 2 times
> totally.  Details are below:
> 
>       C/S of ia64-unstable    syscall (lps)
> ----------------------------------------------------
>       cs?????                  95691.4
>       cs????? + (p1-2)        110680.4
>       cs13111 + (p3)          132783.4
>       cs13111 + (p3-4)        168623.8
>       cs13364                 163274.5
>       cs13364 + (p5-10)       170991.6
>       cs13364 + (p5-11)       171355.7
>       cs13364 + (p5-12)       214037.5
>       cs13364 + (p5-13)       216131.5
> 
> The benchmark was run on a machine who has four-core (1.4GHz) and 4GB
> RAM.  The guest had 1 vCPU and 1GB vRAM.  Unfortunately, I forgot the
> C/S number about the first two measurements but it was not too old.
> The parenthesis encloses patch numbers (see below).  It is natural
> that more realistic workload such as SPECjbb did not achieve a
> significant improvement because system call dominated only a part of
> such workload.
> 
> Series of patches:
> p1/  [IA64] vcpu_get_psr fix
> p2/  [IA64] remove the requirement (vpsr.ic==0) from hyperprivop
> p3/  [IA64] Enable fast hyperprivop for ssm.i and rfi
> p4/  [IA64] Use rum psr.be instead of XEN_HYPER_RSM_BE
> p5/  [Xen-ia64-devel]PATCH 1/3]Can't inject event, when guest is
> executing rfi
> p6/  [Xen-ia64-devel][PATCH 2/3] merge increment_iip
> p7/  [Xen-ia64-devel][PATCH 3/3]provide hypercall the same path with
> syscall p8/  RE: [Xen-ia64-devel][PATCH 3/3]provide hypercall the
> same path with syscall
> p9/  RE: [Xen-ia64-devel][PATCH 3/3]provide hypercall the same path
> with syscall
> p10/ RE: [Xen-ia64-devel][PATCH 3/3]provide hypercall the same path
> with syscall
> p11/ [Xen-ia64-devel][PATCH] Simplify lazy cover algorithm
> p12/ [Xen-ia64-devel][PATCH] Implement hyper_get_psr
> p13/ [Xen-ia64-devel]PATCH] Remove duplicate check is_running_on_xen
> 
> Regards,
> Hiroya

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel