Why the Mega Public Clouds Are Unbeatable



  • InfoWorld looks at why the big three cloud providers Amazon, Google and Microsoft are running at a scale that is simply unbeatable. They are driving down computing and storage prices to a point with which no one can compete and doing so while increasing quality and making profits. The bigger they get the better staff that they can hire, the more scale they have.

    Google and Facebook are already fifth and possibly sixth largest server makers in the world. Google is building their own substations to control their power supply! The big three have many of the world's biggest datacenters.

    With the biggest scale allowing the big mega clouds access to better people, equipment, tools, automation, security and profits how can anyone compete?



  • A bit of an old post (2013) but apparently "Facebook doesn't virtualize its servers, because its software already consumes all the hardware resources, meaning virtualization would result in a performance penalty without a gain in efficiency."

    Seems odd, but ok, and I assume their still using Spinning Rust too... (again I know the topic is almost 2 years old)


  • Banned

    @DustinB3403 said:

    A bit of an old post (2013) but apparently "Facebook doesn't virtualize its servers, because its software already consumes all the hardware resources, meaning virtualization would result in a performance penalty without a gain in efficiency."

    Seems odd, but ok, and I assume their still using Spinning Rust too... (again I know the topic is almost 2 years old)

    It's not unusual for cluster computing like facebook and google to still be using physical machines, infact with those kind of work loads you generally don't see any advantage with VMs. VMs are meant to get a lot of servers that themseleves aren't using that much resources at one time together to save money.


  • Service Provider

    @DustinB3403 said:

    A bit of an old post (2013) but apparently "Facebook doesn't virtualize its servers, because its software already consumes all the hardware resources, meaning virtualization would result in a performance penalty without a gain in efficiency."

    Seems odd, but ok, and I assume their still using Spinning Rust too... (again I know the topic is almost 2 years old)

    This is one of those odd situations. I had a 10K compute node (yes, 10K individual servers) cluster on Wall St. We didn't virtualize at the individual server level because the entire thing was treated as a single computer and virtualized at the cluster level.

    This is what Facebook is doing. Those are not really individual computers, the virtualization is up the stack higher.



  • @scottalanmiller Thank you for the clarification on what is likely being done.

    To point it out, I replied in jest as the article I posted mentioned Spinning Rust.


  • Service Provider

    @Jason said:

    VMs are meant to get a lot of servers that themseleves aren't using that much resources at one time together to save money.

    That's consolidation, not virtualization. Virtualization is the abstraction portion, which in turn is good at enabling consolidation but we don't virtualize for consolidation, that's why one to ones are still standard and expected. Consolidation is quite often a huge benefit, but we should be virtualized regardless of that because it offers stability, ease of management, etc.

    Facebook and Google get around this by using custom hardware, making their own drivers, and virtualizing up the stack.



  • @scottalanmiller said:

    @DustinB3403 said:

    A bit of an old post (2013) but apparently "Facebook doesn't virtualize its servers, because its software already consumes all the hardware resources, meaning virtualization would result in a performance penalty without a gain in efficiency."

    Seems odd, but ok, and I assume their still using Spinning Rust too... (again I know the topic is almost 2 years old)

    This is one of those odd situations. I had a 10K compute node (yes, 10K individual servers) cluster on Wall St. We didn't virtualize at the individual server level because the entire thing was treated as a single computer and virtualized at the cluster level.

    This is what Facebook is doing. Those are not really individual computers, the virtualization is up the stack higher.

    You can even do this at home with Maas and Juju 🙂


  • Service Provider

    @johnhooks said:

    @scottalanmiller said:

    @DustinB3403 said:

    A bit of an old post (2013) but apparently "Facebook doesn't virtualize its servers, because its software already consumes all the hardware resources, meaning virtualization would result in a performance penalty without a gain in efficiency."

    Seems odd, but ok, and I assume their still using Spinning Rust too... (again I know the topic is almost 2 years old)

    This is one of those odd situations. I had a 10K compute node (yes, 10K individual servers) cluster on Wall St. We didn't virtualize at the individual server level because the entire thing was treated as a single computer and virtualized at the cluster level.

    This is what Facebook is doing. Those are not really individual computers, the virtualization is up the stack higher.

    You can even do this at home with Maas and Juju 🙂

    Not quite the same. That's just super efficient management of hardware. It doesn't turn the hardware itself into a cluster that acts like a single computer. It acts, more or less, like an autoprovisioning cloud, but not like virtualization.