Re: [Xen-devel] Re: [PATCH 0 of 12] PV on HVM Xen

[Pasi Kärkkäinen <pasik@xxxxxx>]

On Mon, May 24, 2010 at 06:28:42PM +0100, Stefano Stabellini wrote:
> On Tue, 18 May 2010, Christian Tramnitz wrote:
> > Hi Stefano,
> > 
> > what are the particular advantages of running PVonHVM vs. traditional PV 
> > (vs pure HVM)?
> > I'd like to update the wiki with some info about it...
> > 
> 
> Sorry for the late reply, but I wanted to have some numbers to compare
> the three cases...
> 
> PVonHVM guests are as easy to install, use and maintain as pure HVM
> guests but provide performances comparable or even better than
> traditional PV guests.
> 
> In particular you'll find that PVonHVM guests are always faster than
> pure HVM guests, most of the times faster than 64 bit PV guests but
> probably slower than 32 bit PV guests.
> 

Hmm.. are you sure kernbench is a proper benchmark for disk IO? 
kernel compilations fire up a lot of new processes, and that favours HVM guests.

Basicly it seems really weird to me that in the results HVM (without PV 
drivers) 
is faster than pure PV..

If you compare pure disk-IO PV vs. HVM the difference is usually big.. 
PV beating HVM hands down.

-- Pasi

> 
> The following are the results of kernbench run in optimal mode on a 4
> vcpu linux guest with 512MB of ram.
> The host is a Dell Precision T3500 (HAP enabled):
> testbox:~# cat /proc/cpuinfo
> processor       : 0
> vendor_id       : GenuineIntel
> cpu family      : 6
> model           : 26
> model name      : Intel(R) Xeon(R) CPU           E5520  @ 2.27GHz
> stepping        : 5
> cpu MHz         : 2266.806
> cache size      : 8192 KB
> fdiv_bug        : no
> hlt_bug         : no
> f00f_bug        : no
> coma_bug        : no
> fpu             : yes
> fpu_exception   : yes
> cpuid level     : 11
> wp              : yes
> flags           : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat 
> clflush acpi mmx fxsr sse sse2 ss ht nx constant_tsc pni est ssse3 sse4_1 
> sse4_2 popcnt ida
> bogomips        : 4538.00
> clflush size    : 64
> power management:
> 
> processor       : 1
> vendor_id       : GenuineIntel
> cpu family      : 6
> model           : 26
> model name      : Intel(R) Xeon(R) CPU           E5520  @ 2.27GHz
> stepping        : 5
> cpu MHz         : 2266.806
> cache size      : 8192 KB
> fdiv_bug        : no
> hlt_bug         : no
> f00f_bug        : no
> coma_bug        : no
> fpu             : yes
> fpu_exception   : yes
> cpuid level     : 11
> wp              : yes
> flags           : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat 
> clflush acpi mmx fxsr sse sse2 ss ht nx constant_tsc pni est ssse3 sse4_1 
> sse4_2 popcnt ida
> bogomips        : 4538.00
> clflush size    : 64
> power management:
> 
> processor       : 2
> vendor_id       : GenuineIntel
> cpu family      : 6
> model           : 26
> model name      : Intel(R) Xeon(R) CPU           E5520  @ 2.27GHz
> stepping        : 5
> cpu MHz         : 2266.806
> cache size      : 8192 KB
> fdiv_bug        : no
> hlt_bug         : no
> f00f_bug        : no
> coma_bug        : no
> fpu             : yes
> fpu_exception   : yes
> cpuid level     : 11
> wp              : yes
> flags           : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat 
> clflush acpi mmx fxsr sse sse2 ss ht nx constant_tsc pni est ssse3 sse4_1 
> sse4_2 popcnt ida
> bogomips        : 4538.00
> clflush size    : 64
> power management:
> 
> processor       : 3
> vendor_id       : GenuineIntel
> cpu family      : 6
> model           : 26
> model name      : Intel(R) Xeon(R) CPU           E5520  @ 2.27GHz
> stepping        : 5
> cpu MHz         : 2266.806
> cache size      : 8192 KB
> fdiv_bug        : no
> hlt_bug         : no
> f00f_bug        : no
> coma_bug        : no
> fpu             : yes
> fpu_exception   : yes
> cpuid level     : 11
> wp              : yes
> flags           : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat 
> clflush acpi mmx fxsr sse sse2 ss ht nx constant_tsc pni est ssse3 sse4_1 
> sse4_2 popcnt ida
> bogomips        : 4538.00
> clflush size    : 64
> power management:
> 
> 
> 
> 64 bit PV on HVM
> ----------------
> Average Optimal load -j 4 Run (std deviation):
> Elapsed Time 215.307 (10.1294)
> User Time 632.503 (6.4785)
> System Time 115.497 (4.53905)
> Percent CPU 347.333 (15.885)
> Context Switches 43319.7 (2088.39)
> Sleeps 48950 (3140.18)
> 
> 64 bit pure HVM
> ---------------
> Average Optimal load -j 4 Run (std deviation):
> Elapsed Time 235.25 (4.10598)
> User Time 512.48 (1.69714)
> System Time 73.5967 (0.65041)
> Percent CPU 248.667 (4.93288)
> Context Switches 35930.3 (342.837)
> Sleeps 56249.7 (2784.42)
> 
> 64 bit traditional PV
> ---------------------
> Average Optimal load -j 4 Run (std deviation):
> Elapsed Time 248.187 (12.3954)
> User Time 535.283 (0.818189)
> System Time 127.497 (0.342685)
> Percent CPU 266.667 (12.8582)
> Context Switches 32978 (2968.78)
> Sleeps 54391.3 (3141.44)
> 
> 
> the results show that the 64 bit PV on HVM case is the fastest but it is
> also the one that uses most CPU and User time.
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel

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