io performance regression between xen and XCP

io performance regression between xen and XCP

[Xiao Guangrong]

Hi guys,

We have done the io performance test between xen and XCP, both tested on a
separate disk, and found XCP's regression is very great:

XCP's parameters:
Ｄom0
====== Control[d44ecc85-5539-438d-9bbc-f596dec69617] info:
VCPU:
                  VCPUs-params (MRW): mask: 0,1
                     VCPUs-max ( RW): 8
              VCPUs-at-startup ( RW): 8
                  VCPUs-number ( RO): 8
             VCPUs-utilisation (MRO): 
Mem:
                 memory-actual ( RO): 2146172928
                 memory-target ( RO): 2146172928
               memory-overhead ( RO): 1048576
             memory-static-max ( RW): 2146172928
            memory-dynamic-max ( RW): 2146172928
            memory-dynamic-min ( RW): 1887436800
             memory-static-min ( RW): 307232768

Guest:
====== spv1[9a1a2bf2-41a7-ca00-8dc5-92c51b6ed992] info:
VCPU:
                  VCPUs-params (MRW): mask: 7
                     VCPUs-max ( RW): 1
              VCPUs-at-startup ( RW): 1
                  VCPUs-number ( RO): 1
             VCPUs-utilisation (MRO): 0: 0.000
Mem:
                 memory-actual ( RO): 1073741824
                 memory-target ( RO): 1073741824
               memory-overhead ( RO): 1048576
             memory-static-max ( RW): 1073741824
            memory-dynamic-max ( RW): 1073741824
            memory-dynamic-min ( RW): 1073741824
             memory-static-min ( RW): 1073741824


Xen's parameters:

###### xm list ######
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0  2048     2     r-----     52.5
co5.4-32-2                             	     1  1024     1     -b----     33.0
###### xm vcpu-list ######
Name                                ID  VCPU   CPU State   Time(s) CPU Affinity
Domain-0                             0     0     0   r--      42.2 0
Domain-0                             0     1     1   -b-      10.4 1
co5.4-32-2                           1     0     6   -b-      33.0 6
###### xm sched-credit ######
Name                                ID Weight  Cap
Domain-0                             0    256    0
co5.4-32-2                           1    256    0


Test result:

- direct read/write test:
test command:
dd if=/dev/zero of=1.img oflag=direct bs=256k count=4096
dd if=1.img of=/dev/null iflag=direct bs=256k count=4096

Xen: write: 33.3 MB/s read: 206 MB/s
XCP: write: 18.5 MB/s read: 108 MB/s

- no direct read/write test:
test command:
dd if=/dev/zero of=1.img  bs=256k count=4096
dd if=1.img of=/dev/null  bs=256k count=4096

Xen: write: 319 MB/s read: 85.9 MB/s
XCP: write: 136 MB/s read: 63.8 MB/s

Now, we are very confused about this result :-(


Xiao

Re: [Xen-devel] io performance regression between xen and XCP

[Daniel Stodden]

On Mon, 2010-05-31 at 13:58 -0400, Xiao Guangrong wrote:
> Hi guys,
> 
> We have done the io performance test between xen and XCP, both tested on a
> separate disk, and found XCP's regression is very great:
> 
> XCP's parameters:
> Ｄom0
> ====== Control[d44ecc85-5539-438d-9bbc-f596dec69617] info:
> VCPU:
>                   VCPUs-params (MRW): mask: 0,1
>                      VCPUs-max ( RW): 8
>               VCPUs-at-startup ( RW): 8
>                   VCPUs-number ( RO): 8
>              VCPUs-utilisation (MRO): 
> Mem:
>                  memory-actual ( RO): 2146172928
>                  memory-target ( RO): 2146172928
>                memory-overhead ( RO): 1048576
>              memory-static-max ( RW): 2146172928
>             memory-dynamic-max ( RW): 2146172928
>             memory-dynamic-min ( RW): 1887436800
>              memory-static-min ( RW): 307232768
> 
> Guest:
> ====== spv1[9a1a2bf2-41a7-ca00-8dc5-92c51b6ed992] info:
> VCPU:
>                   VCPUs-params (MRW): mask: 7
>                      VCPUs-max ( RW): 1
>               VCPUs-at-startup ( RW): 1
>                   VCPUs-number ( RO): 1
>              VCPUs-utilisation (MRO): 0: 0.000
> Mem:
>                  memory-actual ( RO): 1073741824
>                  memory-target ( RO): 1073741824
>                memory-overhead ( RO): 1048576
>              memory-static-max ( RW): 1073741824
>             memory-dynamic-max ( RW): 1073741824
>             memory-dynamic-min ( RW): 1073741824
>              memory-static-min ( RW): 1073741824
> 
> 
> Xen's parameters:
> 
> ###### xm list ######
> Name                                        ID   Mem VCPUs      State   Time(s)
> Domain-0                                     0  2048     2     r-----     52.5
> co5.4-32-2                             	     1  1024     1     -b----     33.0
> ###### xm vcpu-list ######
> Name                                ID  VCPU   CPU State   Time(s) CPU Affinity
> Domain-0                             0     0     0   r--      42.2 0
> Domain-0                             0     1     1   -b-      10.4 1
> co5.4-32-2                           1     0     6   -b-      33.0 6
> ###### xm sched-credit ######
> Name                                ID Weight  Cap
> Domain-0                             0    256    0
> co5.4-32-2                           1    256    0
> 
> 
> Test result:
> 
> - direct read/write test:
> test command:
> dd if=/dev/zero of=1.img oflag=direct bs=256k count=4096
> dd if=1.img of=/dev/null iflag=direct bs=256k count=4096
> 
> Xen: write: 33.3 MB/s read: 206 MB/s
> XCP: write: 18.5 MB/s read: 108 MB/s
> 
> - no direct read/write test:
> test command:
> dd if=/dev/zero of=1.img  bs=256k count=4096
> dd if=1.img of=/dev/null  bs=256k count=4096
> 
> Xen: write: 319 MB/s read: 85.9 MB/s
> XCP: write: 136 MB/s read: 63.8 MB/s
> 
> Now, we are very confused about this result :-(

These are bare FS writes, not guest throughput?

On ext3 with the 2.6.27 kernel?

Try turning off the barrier flag on the root fs.

Daniel