-4

I was trying to do cost comparison between bare metal servers and cloud servers from AWS. I could not exactly find how to compare a bare metal server with AWS server. AWS servers give capacity in vCPU. I was thinking that I would google and find what the comparison between a vCPU and a bare metal server is. I could not find such a thing. vCPU is 'barely' explained - just that one should consider it equivalent to a hyper thread.

Short of running my own tests, are there good results published at some place that I can look at?

If I have to run my own tests, then any recommendation on bench marking tools would be highly appreciated. I searched and the number of tests/benchmarks overwhelmed me. Seems like days long exercise.

-Amit

Amit Agrawal
  • 11
  • 1
  • In first approximation each VCPU equals a CPU core, but the actual number of CPU cores may be oversubscribed (i.e. a physical server with 24 cores might host the equivalent of 96 t2.micro instances, an over subscription factor of 4) but the typical financial business case for Cloud is trading CAPEX for OPEX (you pay for what you actually use, rather then investing in capacity you might need in the future, yada yada ya - But the problem addressed first and foremost by understanding your own workloads and benchmarking and then you decide on a suitable architecture. (cloud or bare-metal) – HBruijn Aug 17 '15 at 07:01

1 Answers1

3

Rather than comparing AWS to bare metal you should consider installing a hypervisor of your own and doing to comparisons between your application running on bare metal, vs it running on a equally sized VM.

Oversubscription of resources does not come into performance comparisons since that is a choice the host makes, not a default way in which virtualisation works...

If you do a 1:1 comparison between say a Linux KVM virtual machine with 4 vCPU and 8GB RAM, vs a physical server with the same hardware (4 cores, 8GB RAM, same generation cpu, same underlying storage) you should find that performance is within about 5% of bare metal.

Large scale virtualisation such as Amazon EC2 is tiered for a reason. If you pay for the smallest instances, performance obviously isn't your primary concern and therefore Amazon will cram as many instances of their bottom tiers onto the same hardware as they can until resources are totally exhausted. Many hosts, not just Amazon will assume that if you need performance, then you will pay for higher tiered VM's or your own dedicated hardware to run VM's on.

tomstephens89
  • 1,011
  • 1
  • 12
  • 24