Talk given at the Southern California Linux Expo in March, 2023. Video: https://www.youtube.com/watch?v=H3YbsOCb4lc
The Forgotten Operator
Oxide Computer Company
In the beginning…
• In the beginning, computers were so expensive that they were shared by
necessity – leading to the rise of a (brief) utility computing movement
• But with Moore’s Law, computing became denser, faster – and cheaper
• With each successive turn – minicomputers, servers, workstations,
personal computers – computing became cheaper and easier to own
• By the 1990s computing was only on-premises
The pain of on-premises compute
• With the rise of the internet, compute needs exploded
• All infrastructure was on-premises – it can’t just be spun up!
• Physical infrastructure is capital and labor intensive
• Adding insult to injury, it was all proprietary – hardware and software
• Physical buildout was exceedingly painful
• A conﬂuence of trends began to give rise to an alternative…
The rise of cloud computing
• Several factors in the 2000s came into conﬂuence:
○ Internet ubiquity + protocol maturity
○ Rise of open source software at all layers of the stack
○ Dominance of x86 + “commodity” hardware
○ Strartup ice age + ﬁnancial crisis (emphasis on opex over capex)
• Added up to cloud computing: shared, elastic, API-driven
Myths of cloud computing
• Cloud computing’s ubiquity in the 2010s gave rise to several myths…
• Myth: Cloud computing is a low margin business
• Reality: Cloud computing is a high margin business!
• Myth: The economies of scale from operating a public cloud primarily
accrue to purchasing power
• Reality: Purchasing power is not unimportant – but the much greater
dividend was the ability to invest in innovation!
Cloud computing divide
• Cloud computing operators – hyperscalers – investied relentlessly in
innovation, yielding an increasing divide
• This innovation drove down their own costs, allowing them to bolster
their own positions and continue to innovate
• On-premises infrastructure providers didn’t understand the cloud, and
increasingly focussed on those customers that shared their confusion
• All of this served to accelerate the demise of on-premises compute
• So… is anyone left on-prem?
The forgotten operator
• There (emphatically!) remain good reasons to run on-prem!
• If you are on-prem in 2023, the reasons are likely good
• These include: risk management, regulatory compliance, latency, and
• This on-premises operator has been forgotten by everyone
○ Vendors don’t understand their use case
○ Fellow technologists act like they have never heard of the cloud!
The pain of the forgotten operator
• The forgotten operator is an extraordinary amount of pain: the
abstractions for on-premises compute remain vestigial
• Power, cooling, BMC, BIOS, ToR switch, all date from the PC era!
• And this is to say nothing of the software!
• These systems operate at cross-purposes: they were never designed
together – and to the contrary
• But do we care?
Why we might care
The mandate for rack-scale machines
• Those repatriating onto on-premises infrastructure will (rightfully) expect
API-driven elastic infrastructure
• However, that’s not what they’re going to ﬁnd
• What they will ﬁnd is, in fact, worse than they might remember
• We believe that we must do better
• We must design rack-scale machines that integrate hardware and
software into a single, software-driven system!
Reasons for optimism
• There are a couple of interesting trends that give optimism…
○ Hardware is easier than ever before – and increasingly open
○ There have been tremendous software advances, e.g. Rust and P4
○ Remote teams make it easier to ramp than ever before
• Still, rack-scale design presents new challenges – and is a big build!
Oxide Computer Company
• We have built a true rack-scale machine,
with integrated hardware and software,
allowing one to easily deploy cloud
• After a three year build (!), we are on the
cusp of shipping our ﬁrst product
• We can now say it unequivocally: the
future demands rack-scale design!
Rescuing the forgotten operator
• The public cloud will always play an important role – it’s not going away
• But more and more operators will need to manage both on-prem and
public cloud buildout
• Those operators have the right to modernity, wherever they deploy!
• To the forgotten operator: help is on the way!
• Join us and learn more at https://oxide.computer – and check out our
weekly Discord, “Oxide and Friends” (now also a podcast!)