Want to compare my ESXi RAM allocation with others

[mikkydoos]

Hi all,

 

I'd like to compare my ESXi RAM allocation for Windows guest VMs with others. (VMWare only please, not Hyper-V)

 

Scenario is... I'm running low on RAM on an ESXi 6.0 hypervisor and I think I'm probably allocating far to much RAM to my guest VM's.

 

Server spec here:

 

 

CPU allocation on all my VMs is next to nothing but RAM is maxing out and rather than just chuck in more, I think I can free up quite a lot.

 

Host memory utilisation is only 450MB.

 

 

Guest VM's are:

 

Server 2012: Sims.net and SQL server 2012, a few data extractors = 16GB allocated

Server 2012: DC, DNS, AD etc = 8GB allocated

Server 2012: Windows deployment, Antivirus console, few other low intensity tasks = 8GB allocated

APC Powerchute Virtual appliance = 0.5GB allocated

Server 2012: Print server / Papercut = 8GB allocated

VCentre Server Virtual Appliance = 10GB allocated - no settings changed from pre-set.

Server 2016: Veeam backup and replication = 12GB allocated

 

Plus a couple of test Windows machines that are powered off - 1 with 4GB, 1 with 2GB.

 

 

I've allocated these amounts because the server machines run mega slow with less than 8GB.

I've given the Veeam guest 12GB because the backups take a while with less.

The Sims.net server was running too slow with 8GB so I doubled it knowing that SQL server will eat as much RAM as it possibly can.

 

Each machine has a single 1Gbit virtual NIC.

Host and datastores are running on local storage - 1TB RAID1 with spare.

 

If I reduce the RAM by anything meaningful on any of them, they run very slowly - e.g. they are very slow to respond when operating them in the console or remote desktop and very slow to boot.

 

Active memory on each of these is reported as less than 2GB per machine.

 

This is the one with 16GB allocated.

 

 

 

I'm reading that people are running these machines on 2GB RAM - I cant!

 

It doesn't seem that there is disk or network bottlenecking.

 

Can I get some comparisons please. If anyone is willing to share I'd be grateful.

 

 

Cheers in advance

Edited June 7, 2017 by mikkydoos

[Domino]

Maybe look at rightsizing your CPU allocation? That one VM listed has half of your logical processors. The reason the utilisation is low may be because the machines actually can't utilise it.

 

if you look at normal performance stats for that machine, does the cpu use sit under 50% if so, then halve it, and all the others, you can see the impact it's having using the "cpu-ready" metric in the performance stats. By consuming CPU time with idle processors you're not only making that machine wait for available slots, but you're also denying those slots to other machines. More here: https://blogs.vmware.com/vsphere/2013/02/the-performance-cost-of-smp-the-reason-for-rightsizing.html and here: VM Right Sizing – An example of the benefits | CloudXC

 

I'd suggest looking at that first.

[3s-gtech]

For my ESXi 5.5 VMs

 

DC with AD, DNS, DHCP etc (Server 2012 R2) - 4GB

Print server (Papercut also installed) - Server 2008 32 bit - 4GB

SIMS - Server 2008 R2 - 16GB

Sophos Antivirus console - Server 2008 32 bit - 2GB (below the recommended but runs fine)

 

I think you can trim up a bit on your DC and print server. All of the above work absolutely fine with their allocations, and we're a good sized school.

 

I should add to this by saying that this host runs five VMs with 30GB total allocated (out of 32GB). All fixed allocation. Two RAID 10 arrays - one 10k SAS, the other 120GB Intel SSDs (for SIMS). I made sure that SIMS doesn't touch the disks used by the other VMs due to its heavy utilisation on startup, but SQL is confured to use a minimum of 12GB so it loads the whole DBs into RAM at all times.

Edited June 7, 2017 by 3s-gtech

[mikkydoos]

@Domino

 

For that machine.....Way way way under....

 

 

 

So are all the others.

 

 

I've read that the VM CPU should reflect the number of physical cores - looking at right sizing thats.. wrong.

 

Thanks @Domino I'll take a good look at that.

Edited June 7, 2017 by mikkydoos

[Domino]

@Domino

 

For that machine.....Way way way under....

 

[ATTACH=CONFIG]43678[/ATTACH]

 

 

So are all the others.

 

 

I've read that the VM CPU should reflect the number of physical cores - looking at right sizing thats.. wrong.

 

Thanks @Domino I'll take a good look at that.

 

I'm not sure if the edition you're using is different, but I'd suggest if you've historical data available to look at the load over a period of time

 

[Blue_Cookeh]

You ought to be able to cut a lot of those Windows Server VMs by half for RAM. Particularly printing and DCs.

 

What disks are you running on? Storage is the primary reason for VM infrastructure to feel sluggish.

 

EDIT: I started deploying Server Core with 2016, too. I now allocate 2GB RAM to roles like DC.

Edited June 7, 2017 by Blue_Cookeh

[Theblacksheep]

Ram looks OK, I usually have the DC running 4gb.

 

Check your CPU 'ready time' (in MS) for each VM.

[pete]

DCs are usually 2-4GB here. They start on 4GB and then we check what they're using after a month and reduce if appropriate.

[mikkydoos]

Thanks all. Thats really good advice. Very useful to compare.

 

------------ @Domino

 

Big big thanks. I carefully read those links you posted. I think you are spot on. I've over committed the CPU and there is/was obviously an access issue on the CPU.

I had read (and set up some machines) that VMs should each represent the number of physical CPU core but thinking about it - thats a load of tosh.

 

On a few VMs I've reduced the cores to a single core/socket, reduced the RAM right down and they are running much much better. MILES faster than before.

 

Obviously I'll be keeping an eye on the stats to see if they are under specced now, but for now. That has more than improved them and I have halved the total RAM I was using.

 

Big thanks and I owe you a virtual beer m8

 

 

 

-------------------

 

For the others, have a good read of that. Really helpful

Edited June 7, 2017 by mikkydoos

[Domino]

Nice one dude, glad you're sorted.

 

I had a hell of a job here explaining to dev teams that removing CPUs would make things faster - count yourself lucky that you didn't need to do that

[mrnoisy]

Glad your sorted, in your initial post you indicated you are running Veeam, do you have Veeam reporter ? It's very good at capacity planning and showing you where you have over provisioned vm's, we clawed back lots of resource and improved performance by looking at these reports.

[mikkydoos]

Glad your sorted, in your initial post you indicated you are running Veeam, do you have Veeam reporter ? It's very good at capacity planning and showing you where you have over provisioned vm's, we clawed back lots of resource and improved performance by looking at these reports.

 

No. Just the standard Veeam B&R. Will have a looksy though

[mikkydoos]

So just Veeam on server 2016 doesn't like being a single core CPU and is eating 6gb of RAM.

 

Didn't think it was so bloaty.

[mikkydoos]

Can I ask you guy for your specs of a VM running Veeam B&R if you have.

 

Mines absolutely eating CPU.

 

1 socket, 2 cores - 3250Mhz used !

 

6Gb RAM maxed.

 

Playing with it now but wouldn't mind some comparisons.

 

Cheers again guys

[mikkydoos]

Can I ask you guy for your specs of a VM running Veeam B&R if you have.

 

Mines absolutely eating CPU.

 

1 socket, 2 cores - 3250Mhz used !

 

6Gb RAM maxed.

 

Playing with it now but wouldn't mind some comparisons.

 

Cheers again guys

 

 

Hhhmmm. Seem to have found a sweet spot on this one. 2 CPUs/1 socket, 8Gb RAM.

 

Any less and it just doesn't have it.

 

 

Thanks again all