I think a good design should be reflective of what business goals you are trying to achieve. A lot of times in the technology field designs decisions are made to take into account limitations of certain technologies first and then the business goals. One such design decision is how many VM’s you’re going to fit onto on a LUN or volume.
If your a fiber channel or iSCSI shop you’re going to be using VMFS for your file system of choice when running vSphere. The max number of VM’s is around the 128 mark with a bit higher or a bit less if your vendor of choice supports the atomic test and set (ATS) algorithm to lock the LUN or not. In VDI and Cloud environments the number of VM’s can go far north of 128 for the whole environment.
The issue than becomes managing all of the LUNS for your environment and operations best practices needed for when the storage subsystem needs to grow. If we settle on 128 being the Max number we aren’t going to design for that number, you’re probably going to shoot for 80 so you have some room for growth. Below is representation of how my LUNS you will need to host 1,000 VMS. Looks fun right? Add another 1,000 VM’s and your problem multiplies. Zoneing, Masking, CHAP, Radius, rescanning are some the fun tasks you will have before you.
Traditional NFS is a bit better. 256 VM’s seems like the magic number before performance problems hit with traditional storage controller architecture. Even the 256 limit gets into lots of volumes to maintain and monitor so you don’t cross the threshold.
The Nutanix Distributed File System has no such limits. Distributed storage controllers and all the writes happening locally to the VM prevent the number of VM’s from causing a performance bottleneck if you have the available IO for VM. Since Nutanix nodes scale compute and IO in a linear fashion, capacity planning is greatly reduced.
Building a system is out is usually the easy part too, maintaining a system is the hard part. What happens when someone new or a project gets passed to the operate team and they haven’t been briefed on all the nuisances. More VM’s get deployed and the performance goes belly up. I have seen this happen more than once, unfortunately in my life I have also have been the cause. With the add node feature of Nutanix, two clicks and more storage is added to the existing volumes. There is no need to change your design or add new volumes because of imposed limitations.
As IT shops will all are competing with the Amazon’s of the world. What can you get your admin to VM’s ratio to? Amazon is around the 1 to 10,000 mark. Another certainty for most shops is that you aren’t going to able to get more staff. Budgets and lack of available resources are usually the underlying issue for this.
How may LUNS\Volume do you want to maintain?