An Interview with
The co-founder of VMware and Associate Professor of Computer Science
at Stanford explains how an academic research project evolved into
a commercial idea.
By Asaf Cidon and Tomer London
ASAF CIDON AND TOMER LONDON:
what is the academic lineage of vmware?
MENDEL ROSENBLUM: The academic roots
of VMware date back to a group that led
the d ASH and FLASH multiprocessors,
which were machines with a very [large]
number of processors (up to 4,000) in
a shared memory configuration. our
responsibility in the project was to create
a novel operating system to manage
their computation and shared memory.
The research we were conducting was
effectively an operating system for a
supercomputer. However, I wasn’t really
interested in supercomputers or high-performance computing.
Instead, what really motivated me was
to understand how operating systems
can scale to thousands of processors.
So I made up my own motivation for
our project: What if you wanted to run
a lot of individual people’s machines
or applications on a big centralized
compute engine. That drove us to take
design decisions that weren’t exactly
needed for high-performance computing.
For example, we looked a lot at how to
make the machine resilient to failures,
e.g. if one processor failed, the rest of
the machine could keep running. This
was a very unusual feature of shared
memory machines at the time, but it was
a very interesting challenge for our own
motivation. our research achieved pretty
good success academically.
The practical concern with our
research was that the commercial
marketplace wasn’t ready for such
solutions. First, nobody had really built
machines at the scale of dASH or FLASH.
Second, it wasn’t really an option for
UnIX or Windows to accommodate the
massive changes required to scale to
thousands of processors.
I ended up giving a talk at Sequent
Computer Systems, which had a team
developing an operating system for
multiprocessors. The people at Sequent
told us that our ideas were interesting,
but that their team was too small to
implement our ideas and that they
eventually would want to run Windows.