Hi,

Thanks to Chip and Alex. To test theories, we can use different solutions such 
as random process, the Monte Carlo simulation, and so forth, and we do not even 
always need to pay this much attention to details of a specific structure like 
the Marvin testing page does for it. 
However, by using simulation, I meant something like the solution that has been 
mentioned by Gulatietal. in [1]. They mentioned that: 


"To develop and evaluate various algorithms, we implemented a simulator that 
simulates a cluster of ESX hosts and VMs. The simulator allows us to create 
different VM and host profiles in order to experiment with different 
configurations. For example, a VM can be defined in
terms of a number of virtual CPUs (vCPUs), configured CPU (in MHz) and 
configured memory size (in MB). Similarly a host can be defined
using parameters such as number of physical cores, CPU (MHz) per core, total 
memory size, power consumption when idle etc.
In terms of workload, the simulator supports arbitrary workload specifications 
for each VM over time and generates CPU and memory demand for that VM based on 
the specification."


So, I was wondering if there is something like this available, but it seems 
there is not. 




--------[1] Ajay Gulatietal., "VMware Distributed Resource Management: Design, 
Implementation, and Lessons Learned", 2012


Thanks,

--Pouya

 
> Consider using the simulator plugin, which will simulate hosts.  You can 
> follow
> the testing instructions on the Marvin testing page [1], but modify your setup
> to implement whatever number of hosts you want to test against.  From
> there, you would need to modify the allocator that's enabled.
> 
+1  That's exactly how it should be done.  It's how stress tests are performed.

You can modify the simulator plugin to vary its response to play devil's 
advocate to test out different theories.

Also, the simulator today does not have this but you should be able to make it 
implement the PluggableService interface to expose commands to allow your test 
client to adjust responses on the fly.

--Alex

Reply via email to