Analytics VM memory : 9Gb recommended, less than 4Gb used

[david_loup]

Hello,

When I installed Vcops, I followed the recommendation for the analytics vm : Si it has 9Gb of memory.

My host is running out of memory so I'm trying to find oversized VM, and with Vcops, I can see that the analytics vm never used more than 4Gb.

I saw in the installation guide that the recommendation was 9Gb for up to 1500 VM to monitor. I have a little infrastructure with less than 70 VM.

I will add some in the futur, but I don't think i'll reach 130 VM before years.

So my question is, would it be reasonable to allocate only 4Gb to this VM ?

Thank you in advance.

[ShawnPD]

Hey David,

     I am no expert by any means but I usually take an incremental approach when adjusting resources...I will usually slice off a percentage at time, re-analyze then adjust further.  Make sure you take into account spikes of activity to ensure you do not create constrains.

     My personal experience, we initially set up the analytics vm with 6 GB of RAM until we went over the 1500 vm threshold, then we adjusted to 9 GB of RAM.  We did not see any user experience degredation.

     Good luck!

[CedricAnto]

Hi David,

Here are couple of things to consider

1- Oversized & Undersized VM's are based on policies defined- Unfortunately if you take the default policies they tend to be really aggressive and drag majority of workloads reflecting as wrongly sized. This behaviour has been refined in newer releases wherein you have some template policies. Policies are much like a workloads profile i.e. some VM's are accessed and busy during business hours, some VM's (example backup VMs) are busy during off business hours, similarly some are web based, iops based. Analytics VM also performs some staggered tasks and well as continuous tasks(data collection/processing)

In a nutshell you have to profile VM's and split them into Intelligent groups and apply policies that are applicable for specific groups(type of VMs). Otherwise these typically end up as incorrectly sized in the reports

2-Always respect minimum requirements of applications, natively VCOPS is not application aware, although you can integrate hyperic, it still does not connect the dots to respect application requirements.

For example assume you have a vCenter VM with Tomcat services configured to used 4 GB, but based on VCOPS report of VCenter VM being Oversized, if you reduce vC VM size that cuts into Tomcat applications configuration, that application would crash or perform poorly, This holds good for all applications, so Analytics VM by design sized for minimum 1500 VM's, although you run much smaller environment , reducing the memory is not recommended.(You can ofcourse reduce disk size)

Reference http://kb.vmware.com/kb/2057607

Let me know if this provides atleast a high level perspective/context and answer to your question. Otherwise I will try to elaborate further.