Recently Thomas Dullien wrote a blogpost
<http://addxorrol.blogspot.com/2020/07/the-missing-os.html> asking what the
OS of the future really looks like, considering the computer of the future
is a distributed mega-engine. I would, annoyingly, posit that the
algorithms that make sense to understand in that world are those already
implemented in the many species of social insects.
In that sense, I think there are things missing from his list of the basics
of what a datacenter computer contains. At some level, I think the right
set of algorithmic support will undergo phase changes as you scale up, but
still be viable at small scales: (see this new paper for a detailed
I will include Thomas's list below:
- Some form of cluster-wide file system.
- A horizontally scalable key-value store.
- A distributed consistent key-value store.
- Some sort of pub/sub message queuing system.
- A job scheduler / container orchestrator.
I would say some level of ID is missing - large-colony ants have Caste
systems for a reason. I think people in general ignore ID because maybe you
can make ID and security a micro-service on top of a pub/sub message
queuing system? I'm not sure. But in general, the same thing is true for
most security analysis systems sold today - they see an infection, they
treat a computer/network as infected and respond appropriately.
At Immunity/AppGate we've been trying to go the other direction - tying
your security directly to your identity. AppGate SDP already does this for
network services, but we've also expanded upon that by using a graph
database, INNUENDO Agent, a custom Windows Kernel module, and a lot of
other excitement to build automated response that looks at your ID as much
as it does your computer:
CONSTELLATION + AppGate SDP (3m)
CONSTELLATION vs. Ransomware (9m)
CONSTELLATION vs. Robbinhood (11m)