A Case Study of Hardware and Operating Systems Co-Evolution
Michael Engel - Universität Bamberg

Pessimism in the innovative power of operating systems research was at
the center of a number of discussions over the last decades. Already in
the year 2000, Rob Pike stated in his polemic "Systems Software Research
is Irrelevant'' that our field is in decline. Later, Timothy Roscoe demanded in
his OSDI 2021 keynote that "It's Time for Operating Systems to Rediscover 
Hardware'', since operating systems have become only a small niche topic 
for major OS conferences, not even meriting their own full session.

However, we do not think that the field of operating systems research is in 
decline. One observation to be made, though, is that the feasibility of certain
approaches depends on the evolution of hardware features and properties.
Though some researchers half-seriously state that computer scientists tend to
reinvent the wheel every two decades, the feasibility of implementing certain 
pivotal ideas depends on the relation of non-functional parameters of underlying 
hardware platforms. In turn, we think that re-examining old ideas is worthwhile 
when new technologies on the hardware level arise.

In this talk, we examine this interdependence of hardware and operating system 
evolution by examining a specific use case, the influence of parameters such as 
memory and I/O latency and throughput on the feasibility of OS features. Here, we 
analyze the history of more than three decades of approaches to handle memory 
overcommitment, specifically on-demand memory compression and network-
distributed memory, both of which are important building blocks for the current
trend of disaggregated memories in the context of novel memory hierarchies
including disruptive memory technologies.