Their comparison sizes for instances versus physical machines is laughable. For a 12 core physical machine with 64GB of RAM they list the r3.2xlarge instance size as unit-for-unit comparable (10 instances versus 10 machines I had configured).
Yeah... no. 8 vCPUs are nowhere near as fast as a pair of 6 core Xeons. My rule of thumb in AWS is that you get about half of what they say you do: so an 8 vCPU instance has about as much power as a 4 core physical server.
And don't even get me started on what they have for hardware costs. I think they must have pulled list price from Dell or something. I can get prices about 2/3 to 1/2 of what they're listing from my VAR.
This doesn't seem useful if you are pricing out anything less than a full (or multiple) racks of servers.
For 2 servers with moderate specs (8 cores total, 16gb each) it claims the 3 year cost will be $120K. Most of that seems to be due to the assumption that 2U requires a full rack costing $30K up front plus $2.5K/mo in datacenter costs.
They conveniently leave out the 'real' competition which is 50%+ utilization on dedicated server providers. Consistent performance, significantly cheaper.
I do believe that's the market they're going after. Most people don't calculate TCO unless they're an enterprise looking at buying/managing their own gear, which would include racks.
Obviously completely biased. Well utilised hardware or dedicated servers can easily compete (and blow out of the water at cheaper DC's) on price with AWS (and then you actually get guaranteed resources).
Thanks for that. Nice that the costs are getting closer, I only have to pay an extra million dollars a year to host my crawling and indexing infrastructure in the cloud than to co-locate it by their calc. It would still be impossible to get the latency I want but that the hardware cost difference is down to around 20% is a huge improvement.
Actually the calculator can do that, but badly. You need to go in and delete out the 'rack' cost which is typically built in with most Colocation deals.
The pricing is still waaaaaaaaaaaaay off. They intentionally left it out because anyone with high utilization would find the dedicated servers cheaper.
Yeah... no. 8 vCPUs are nowhere near as fast as a pair of 6 core Xeons. My rule of thumb in AWS is that you get about half of what they say you do: so an 8 vCPU instance has about as much power as a 4 core physical server.
And don't even get me started on what they have for hardware costs. I think they must have pulled list price from Dell or something. I can get prices about 2/3 to 1/2 of what they're listing from my VAR.