Thanks for this very informative and somewhat scary information, very pertinent given our use of tomcat and VMs
Any idea which VMware product this is? vmware server and esx are entirely different software offerings, vmware server is not highly regarded by our infrastructure teams, while esx is essentially a hypervisor model like xen
>-----Original Message-----
>From: Institutional portal discussion list
>[mailto:[log in to unmask]] On Behalf Of Paul Browning
>Sent: 12 June 2009 11:10
>To: [log in to unmask]
>Subject: Re: Confluence performance on a virtualised server
>
>David Rischmiller wrote:
>> Firstly I know nothing about Confluence, but have some experience of
>> running virtual infrastructures based on VMware.
>>
>> That difference in performance suggest to me that you are somehow not
>> comparing like with like. The difference between physical and virtual
>> performance obviously depends on the resource requirements of the
>> principal application involved on the server. If these are being
>> adequately provided (CPU, RAM, I/O and Network bandwidth) then any
>> performance degradation should be very small; definitely not 60:1
>> degradation.
>>
>> There is, no doubt, scope for pathological applications that go
>against
>> all expectations but I think that results in a question mark over the
>> quality of the application and it's unlikely to be the situation here.
>
>Thanks for your response.
>
>Also heard from Steve Daniels @ Edge Hill:
>
>> We run Confluence on a VMWare hosted virtual machine and recently had
>> some performance issues which sparked a discussion between myself and
>> the manager of Core Services. He recalled reading somewhere that if
>> Tomcat can see it has 4 virtual cores it waits for all 4 "cores" to be
>> available before using them all at once. On a dedicated system this is
>> fine, but when those 4 cores/threads are actually spread over numerous
>> dual core processors, which also have about 15-20 odd other virtual
>> machines running on them the delay can give some performance issues.
>>
>> We already had the virtual machine's RAM as real dedicated RAM not
>> virtual. Now we've gone from 4 virtual cores down to 2, but double
>> their processing speed. Starting confluence up went down from a
>> similar 7 minutes or so to a much shorter time.
>>
>
>We're following this up.
>
>Paul
|