In the near future we will see the next generation processors (eight and twelve cores) from AMD and Intel, and shortly afterward an ever growing nuber of servers will be equipped with them. It is likely that computational power, memory and network capabilities will soar to a new level and we will be able to put more Virtual machines on a single physical server.
Now a single 2 cpu server can sport something like 30 VMs, then we will be able to stuff 50/60 VMs in the same hw footprint (more and more if we consider VDI)! WOW! And this numbers will scale up linearly if we think about 4 and 8 CPUs servers!
Increasing number of server for the same footprint have a lot of advantages:
- same costs of licenses (vmware);
- less acquisition costs for hardware;
- less hardware to maintain;
- less connectivity to maintain;
- a simpler overall architecture;
- less power and cooling;
- and so on…
But there is a little thought that comes to my mind: what will happen when a fail accours to one or more of this “nexgen” host servers??? Probably all the VMs will be rebooted on other physical hardware at the same time originating a boot storm on the SAN.
Each on of the VM’s operating systems will perform file systems checks, boot of system and an applications recovery! (needless that some of them will not start properly!!!)
Too many VMs starting at the same time can cause a degradation in the service level for the other VMs, huge response times and, sometime, application problems!
The more guest will start from crash the more problems you will experience during the boot phase!!! Is your IT staff ready to solve problem caused by a flood of VMs stopped at the same time?
So, Are we really ready for these new processors?
Are our networks ready (Data Center Ethernet is still a work in progress, probably inifiniband can fit better) to support tons of VMs in a single node?
Are our SANs ready to support tens of parallel file system checks at the same time plus the normal workload?
Obviously this is a provocation: a better design of server farms will avoid a lot of these problems… but you need to think about it before it will happen!
SMBs will impact more than big guns, since the former the number, thus the cost of physical servers always deserves a leading role in their projects.
Comments are welcome!