Benefits of VMware – the uneven cluster

VMware are often focussing on the latest and greatest features and capabilities offered by their newest software. Of course, they are always driving forward and the next version’s enhancements and benefits are forefront of their minds – but there are still some people out there who are just starting on their virtualisation journey. The advantages offered by the premium versions of vSphere such as Enterprise Plus, offer exceptional advances for businesses and enterprises, but some smaller businesses are unable to afford these editions – particularly at the start.

Some benefits of virtualisation, particularly with vSphere, are inherent and included in all versions – and deliver significant savings in both money and time. There are some the simple benefits that are often not highlighted to new users of virtualisation, but well known to existing users.

It is definitely a trend within the hardware industry to develop servers that are optimised for virtualisation. High memory density, multiple built in network cards, support for the latest multi-core CPUs – and many other enhancements. It’s actually quite hard to buy a good quality server that has just 2 CPU cores and 4 GB of RAM and 40GB of fault-tolerant disk space and a single network card – often this is the requirements of server software for small and medium businesses.

How much is your VM really using?

Interestingly, in my experience of working with VMware customers (and my history of being a VMware customer for 5 years too), some server software actually consumes less resources than even that! It’s common to see a Windows server actively using less than 384Mb of RAM, 200 MHz (that’s 0.2 GHz) of CPU and, after the installation of Windows, less than 8GB of disk space. Why don’t you try and buy a physical server with those specs – you can’t! In discussions that I have had with software vendors, they often “bump up” their official minimum hardware specifications to the level of a mainstream standard server, because customers keep on contacting them to ask if their ‘powerful’ server is appropriate or overkill.

Based on proper analysis (such as through vRealize Operations Manager – vROps), or even careful manual ad-hoc analysis of the vCenter performance statistics over a reasonable time, it might be apparent that your servers are over-sized. The recommendation might come back from vROps that your servers should have 400Mb of RAM, or 3 CPUs. Unusual sizes? Not with vSphere. You can set odd numbers of CPUs (uneven, not ‘strange’…) and memory sizes in increments less than 1GB.

Resizing your VMs – down

There is a tiny drawback here though. If you need to resize your VMs downwards, the operating system would get quite a bit upset if you pulled out memory or CPUs whilst they were in use – that’s why you are prevented from doing it in the vSphere client. Instead, you need to power down the VM, make the changes, and then when it powers on, the new specifications take effect.

The upside is, if you have vSphere 5.1 Standard or above, you have hot-add of CPU and memory (vSphere 4.x and 5.0 needs Enterprise or above). This needs to be activated on each VM (version 7 or later) whilst it is powered off (so we recommend you set this on your templates), and in usual VMware fashion this is a single mouse click GUI option to ‘enable’. Depending on your Windows edition, the memory will be immediately accessible (2003 Enterprise, 2008 Enterprise, 2008R2, and above), and the CPU will be immediately accessible (2008 R2 Enterprise, 2012 and later), or require a reboot (for RAM; 2008 Standard, for CPU; all 2008 and 2008 R2 Standard). For Linux flavours, hot add varies depending on your distro – some recognise the new hardware immediately and some require kernel commands to recognise the additional CPU(s), or may require a reboot.

How big?

How much can you add? It depends upon your license – for vCPU per VM from 8 in the Essentials all the way up to 128 vCPUs per VM in Enterprise Plus (6.0). Memory can be added up to 6TB (actually 6128 GB – in 6.5). However – you can’t give an individual VM more virtual CPUs than you have physical CPU cores, and you can’t add more RAM to a VM than you physically have inside the host server. However, as I mentioned above, smaller specifications are often what is needed in most applications used in medium and smaller businesses.

If you are new to virtualisation, or have never tried running a virtual machine with an “uneven” virtual hardware configuration – give it a try and prove to yourself (and your colleagues) that servers do not always need to have 2 or 4 or 6 (CPUs or GB of RAM). Next time you are purchasing software that has a minimum requirement, have a look at your trial or evaluation of the software and see what it is actually using – this might be a start to a whole new density improvement in your vSphere environment.

If you have got this far and are thinking, “what about large enterprises?” then you need to consider the overall density that sizing your VMs correctly can achieve. If you have a density of around 30 VMs per host, then even reducing each VM by 100MB can release (in this example) a further 3GB. So, “right-sizing” your VMs to uneven numbers can be effective efficient for the VMs, and good for the hosts too!

Uneven Hosts in clusters

So, you can re-size your VMs to be a more appropriate size – rightsizing the VMs to meet the needs. What about if you want to make a cluster that has different sized hosts? With VMware, this is absolutely possible, where you can have a cluster that contains mis-matched hosts – because everything is abstracted from the underlying hardware. There may be reasons not to do this (such as EVC migrating VMs onto a slower host), and you may instead want to have a dedicated management cluster, but you can have a vSphere cluster that is made up of different sized hosts – they just all need to be able to access the same networks and storage, and all have the same class of CPU.

For smaller outfits that have only a few servers, virtualisation can still offer benefits for you that you can achieve without having to spend tens of thousands on a brand new cluster.