Seeking input on specs for VMs to run JMeter

[Scott Dean <ps...@ncsu.edu>]

Greetings!

I have been tasked with transitioning our load testing infrastructure from
physical Windows servers to virtual machines running linux (Redhat
Enterprise Linux 7, to be specific) hosted on VMware.  I am looking for
input on how to spec the VMs in terms of number of vCPUs and RAM as well as
number of VMs.  Would I be better off with fewer beefy VMs or more VMs of
more modest specification?  At present, we have 5 physical servers where
one machine is used as a master/controller and the other four are remote
servers that run the test plans and generate load.  Should the master
server be spec'ed differently from the remote servers?

In the end, I need to be able to make a request to the department
responsible for the VMware environment for a number of VMs with general
specifications regarding # of virtual CPU, allocated RAM, and allocated
disk space.  The hosts in the VMware farm appear to be running Intel Xeon
E7-4860 v2 (Sandy Bridge) @ 2.6GHz CPUs

Background on how we use JMeter:

I work for a large university, with about 35K students and 10K
faculty/staff, and we are using JMeter to load test our Oracle PeopleSoft
ERP systems (HR, Finance, & Student Administration), when we do major
software or hardware upgrades. We transitioned from HP Loadrunner to JMeter
about 2 years ago with JMeter 3.x.  When load testing, our objectives are
to verify that our server/database configurations are sufficient to handle
the anticipated user load for events like student registration, housing
sign-up, or employee annual benefits enrollment, as well as testing for
reasonable response time for particular business processes, like initial
user SSO login to the integrated ERP system, a student registering for
classes, or generating a voucher to pay for something in the financials
system.

Thanks for the help!

-- Scott Dean
    North Carolina State University
    Raleigh,  NC    USA

Re: Seeking input on specs for VMs to run JMeter

[Kirk Pepperdine <ki...@gmail.com>]

Hi Scott,

The issue we typically see in virtualized environments isn’t so much RAM/CPU as it is networking. I could write a very very long email on this topic but for brevity I’d suggest that if you log local to the VM you’ll need 1 physical network card for each VM running on the machine. If you remotely then all the disk I/O is likely to become network I/O and thus it will be competing for the network stack(s) along side your load test activities. In this world, more slower network cards are actually much better than 1 super fast card.

Recommendation on RAM is make sure you have enough physical RAM to support the VM. You can use current stats to get an estimate as to how big you’ll need to make the VM and then look at VM size to determine how much memory overall you’ll need. From there you can do the maths to sort out how much overall memory you’ll need. As for CPU, the metric to watch is steal time. If that is non-zero then you’re most likely good. If not, you need more hardware.

FWIW, I’ve run JMeter on a rack of Raspberry PIs and it worked just fine. Yes the network capabilities of a PI isn’t great but with a switch I was easily able to scale out the PIs to maintain the required load on the server. It was a fun experiment but the conclusion in this case was that more small was better than fewer big.

As an aside, University of Buffalo is one of my customers and I’ve been involved with their tuning of installation of PeopleSoft. What I found was that the turning recommendations from Oracle were sub-optimal and their support wasn’t very helpful to UoB in helping resolve the issues. Most of the issues did resolve around JVM tuning parameters. DB configurations were never an issue. That said, I believe your a wee bit bigger than UoB so I’d imagine that once you get past the issues I encountered at UoB you’ll start hitting other things that we just never needed to deal with.

Kind regards,
Kirk Pepperdine

> On Apr 12, 2018, at 7:56 AM, Rupesh Garg <ru...@gmail.com> wrote:
> 
> surely. I think we can give some inputs here.
> 
> On Wed, 11 Apr 2018 at 02:06 Scott Dean <ps...@ncsu.edu> wrote:
> 
>> Greetings!
>> 
>> I have been tasked with transitioning our load testing infrastructure from
>> physical Windows servers to virtual machines running linux (Redhat
>> Enterprise Linux 7, to be specific) hosted on VMware.  I am looking for
>> input on how to spec the VMs in terms of number of vCPUs and RAM as well as
>> number of VMs.  Would I be better off with fewer beefy VMs or more VMs of
>> more modest specification?  At present, we have 5 physical servers where
>> one machine is used as a master/controller and the other four are remote
>> servers that run the test plans and generate load.  Should the master
>> server be spec'ed differently from the remote servers?
>> 
>> In the end, I need to be able to make a request to the department
>> responsible for the VMware environment for a number of VMs with general
>> specifications regarding # of virtual CPU, allocated RAM, and allocated
>> disk space.  The hosts in the VMware farm appear to be running Intel Xeon
>> E7-4860 v2 (Sandy Bridge) @ 2.6GHz CPUs
>> 
>> Background on how we use JMeter:
>> 
>> I work for a large university, with about 35K students and 10K
>> faculty/staff, and we are using JMeter to load test our Oracle PeopleSoft
>> ERP systems (HR, Finance, & Student Administration), when we do major
>> software or hardware upgrades. We transitioned from HP Loadrunner to JMeter
>> about 2 years ago with JMeter 3.x.  When load testing, our objectives are
>> to verify that our server/database configurations are sufficient to handle
>> the anticipated user load for events like student registration, housing
>> sign-up, or employee annual benefits enrollment, as well as testing for
>> reasonable response time for particular business processes, like initial
>> user SSO login to the integrated ERP system, a student registering for
>> classes, or generating a voucher to pay for something in the financials
>> system.
>> 
>> Thanks for the help!
>> 
>> -- Scott Dean
>>    North Carolina State University
>>    Raleigh,  NC    USA
>> 
> -- 
> Rupesh Garg
> 9160002744


---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscribe@jmeter.apache.org
For additional commands, e-mail: user-help@jmeter.apache.org

Re: Seeking input on specs for VMs to run JMeter

[Rupesh Garg <ru...@gmail.com>]

surely. I think we can give some inputs here.

On Wed, 11 Apr 2018 at 02:06 Scott Dean <ps...@ncsu.edu> wrote:

> Greetings!
>
> I have been tasked with transitioning our load testing infrastructure from
> physical Windows servers to virtual machines running linux (Redhat
> Enterprise Linux 7, to be specific) hosted on VMware.  I am looking for
> input on how to spec the VMs in terms of number of vCPUs and RAM as well as
> number of VMs.  Would I be better off with fewer beefy VMs or more VMs of
> more modest specification?  At present, we have 5 physical servers where
> one machine is used as a master/controller and the other four are remote
> servers that run the test plans and generate load.  Should the master
> server be spec'ed differently from the remote servers?
>
> In the end, I need to be able to make a request to the department
> responsible for the VMware environment for a number of VMs with general
> specifications regarding # of virtual CPU, allocated RAM, and allocated
> disk space.  The hosts in the VMware farm appear to be running Intel Xeon
> E7-4860 v2 (Sandy Bridge) @ 2.6GHz CPUs
>
> Background on how we use JMeter:
>
> I work for a large university, with about 35K students and 10K
> faculty/staff, and we are using JMeter to load test our Oracle PeopleSoft
> ERP systems (HR, Finance, & Student Administration), when we do major
> software or hardware upgrades. We transitioned from HP Loadrunner to JMeter
> about 2 years ago with JMeter 3.x.  When load testing, our objectives are
> to verify that our server/database configurations are sufficient to handle
> the anticipated user load for events like student registration, housing
> sign-up, or employee annual benefits enrollment, as well as testing for
> reasonable response time for particular business processes, like initial
> user SSO login to the integrated ERP system, a student registering for
> classes, or generating a voucher to pay for something in the financials
> system.
>
> Thanks for the help!
>
> -- Scott Dean
>     North Carolina State University
>     Raleigh,  NC    USA
>
-- 
Rupesh Garg
9160002744