Over-allocating memory for VM guests

Memory management in VMware is much better than other products such as HyperV.  You are correct, you should run into problems.  I have about 20 VM's running on a host with 32GB of RAM.  Each of these machines have more than1.5GB allocated to them.

The only time I run into even a little problem is during backups.  I do host based backup, and depeding on the timing, it does get a little close to the limit of 32GB, BUT that is a night and no one is working anyway.  :o)

You might eventually have a look on the discussions at:
https://www.experts-exchange.com/questions/26259865/esxi-ram-and-cpu-configuration.html?sfQueryTermInfo=1+10+30+alloc+esxi+over

Hope it helps.

VMware has a number of memory management techniques that allow you to overcommit physical RAM while experiencing little or no impact to performance. In ESX(i) 3.5 the two main techniques are Transparent Page Sharing (TPS) and Memory Ballooning.

TPS works well when many of your guests are basically running the same OS, service pack, and patch levels - say Windows Server 2003. The idea is that many of resident memory pages contain identicle code, and VMware only has to maintain a single copy of the page in physical RAM that is shared by all of the similar virtual machines. Same holds true for several Linux, or other OS types.

Memory ballooning requires VMware tools to be installed into your guest virtual machiens. The idea is from within the guest the memory driver in VMware tools understands better the needs of the virtual machine, and is unable to release little used memory pages to ESX to be better utilized elsewhere.

So far a a general rule of thumb that is hard to define. As you can see it depends greatly on how similar your workloads within the VMs are. What you can do is from within the VI Client look at the performance tab, select the memory counters from the dropdown. Whenever you see the "Swap" counter come off of zero that is the point where you may begin to experience some performance impact.

Hope this helps

I should proofread - make that "and is ABLE to release little used memory pages..."

All in the PDF link I posted but bgoering said it much better than I did.  I would even say he put it better than VMware!  Hahaha

Whenever you consider RAM in the virtual world, you also have to take overhead into consideration. The VMKernel needs some memory to manage the VMs which it considers as worlds, each VM runs within its own process and the VMKernel manages the same.