Performance ESXi 5

[rahinyaqin]

I am doing a research to compare the performance of FTP server between VMWare ESXi and Xen Hypervisor. I use Passmark to obtain the benchmark rating. Higher rating means that it perform better. However, I obtain weird result for VMWare as follow:

Number of Client        Passmark Rating (ESXi)

1                        874.67

2                        892.67

3                        896.60

Compare this result with Xen Hypervisor:

Number of Client        Passmark rating (Xen)

1                        263.19

2                        230.00

3                        184.45

It is quite strange for ESXi since the higher number of Client result on the higher rating. Based on my study, it should be otherwise? Anyone have any idea about this strange result?

[frank_wegner]

Can you share some more details on your test setup? Do you measure inside the VM? If so you need to consider that timing measurements (especially in the millisecond range) do not really provide useful numbers, because of the way the hypervisor schedules the virtual CPUs. A good way to measure performance is to use an external system.

[Josh26]

An FTP server utilises neither CPU nor RAM and is mostly a function of network bandwidth.

Even a very old server should be capable of flooding a NIC on FTP downloads, so this is a fairly terrible performance test. Assuming that rating is some sort of "total throughput", increasing the clients shouldn't have much impact (and it doesn't, those numbers are close enough to consider "the same" IMO) until you start to run 500-1000 clients.

[rahinyaqin]

Here is my details setup on ESXi :

OS                               : Windows Server 2003

Virtual Machine Verison  : 7

Memory Size                 : 1 GB

Network Adapter Type    : E1000

SCSI Controller              : LSI Logic Parallel

Disk size                       : 40 GB

Disk Provisioning            : Thick Provisioning Eager Zeroed

Virtual Device Mode        : IDE

I test performance ESXi inside the VM. Can you give me more details about timing measurements?

And how do you test with an external system?

[rahinyaqin]

That rating is not total throughput but CPU rating. Here is my details result.

Number of Clients                          Passmark rating (ESXi)

                                  CPU            Memory                              Disk                       

1                                874.67         534.40                               486.08

2                                892.67         538.40                               657.90

3                                896.60         540.53                               680.20

Compare this result with Xen Hypervisor:

Number of Clients                           Passmark rating (Xen Hypervisor)

                                  CPU            Memory                              Disk                       

1                                263.19         175.37                                337.92

2                                230.00         165.04                                320.73

3                                184.45          124.96                               311.81

There is no network parameter because Passmark does not provide for measurement real network simulation.

Maybe I made a mistake when I measured performance of FTP server. So, do you have any idea on the best way to test the FTP server?