OT: Rogers and Home-based Business Networks?

[Marc Lanctot]

On 14/04/09 12:13 PM, Robert Brockway wrote:
> On Tue, 14 Apr 2009, Marc Lanctot wrote:
>
>> I have never liked the idea of running a server on a virtual OS. It
>> seems to be the popular thing these days but to me it's always been
>> just a waste of perfectly good hardware.
>
> That's interesting. One of the reasons I'm such a fan of virtualisation
> is that it makes such good use of perfectly good hardware :)
>
> Running a box mostly idle is a waste of hardware and energy, IMHO.
> Running 50 or 100 virtual boxes provides much better use of the hardware.
>
> Advantages include lower purchase costs, lower running costs, task
> seperation so important in prod networks.

Haha, so I should have post-scripted my previous message with this: I 
can see how it works well for web site serving, because given the nature 
of the usage per time of the web server, yes, everything you have up 
there makes sens. And given then probably 90% of servers is doing simple 
web serving with maybe some DB back-end I can see why virtualization is 
so hot these days. Maybe it's smartest way to host web sites.

However, I can assure you what I plan to do will thrash the servers CPU. 
And there will be a lot of inter-networking between my servers going on. 
So in my case, performance per time unit is quite important. I'm not 
sure how the networking between servers would be affected by 
virtualization.. it depends on the setup, probably.

So my concern is if I only get a percentage of the CPU. Say there are 10 
other OS's on my virtual machine. Will my processes ever be allowed to 
get more than 1/10th of the CPU power? If so, let's say my processes are 
hogs and take up 80% of the physical machine's CPU at all times.. will 
this ever be a problem?

>> - Have you ever done any performance tests and compare the result to
>> an equivalent non-virtual OS on the same hardware? Since I'm doing
>> more than web
>
> Lots of testing has been done as this is a key question. Virtualisation
> covers many different products with different characteristics so
> performance testing is very much specific to a particular virtualisation
> app. You can find a lot about virtualisation performance online.
>
> Physical/virtual comparisions also vary a lot based on the workload.
> Some forms of virtualisation run at close to 100% for CPU but are slow
> for disk I/O for example.
>
> I setup OpenVZ[1] at work and it typially performs at upwards of 97% of
> the performance of the physical hardware. OpenVZ writes directly to real
> filesystems so doesn't suffer any I/O performance problems.
>
> [1] Technically this isn't virtualisation at all, it is jailing but
> these two concepts often solve the same problems.

Do you know what virtualization setup they have at Linode? In 
particular, will I *ever* be allowed to use more than 1/Xth of the 
physical CPU?

I will contact them and get back to the list.

Marc

-- 
Imagination is more important than knowledge.
   -- Albert Einstein
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists