Where is compute (physically) headed in the near future?
Where is compute (physically)
headed in the near future?
History of computation as oscillation
• The history of computation is one of repeated oscillation between
centralization and decentralization
• These oscillations are driven by economics:
• Economies-of-scale drive towards centralization
• Disruptive innovations drive towards decentralization
• These oscillations can also be seen as the tension between
control and freedom — with each having economic advantages
with respect to the other
Cloud computing as centralizing force
• By the 1960s, pundits foresaw an ultimate centralization: a
compute utility that would be public and multi-tenant
• The vision was four decades too early: it took the internet +
virtualization + commodity compute to yield cloud computing
• Public cloud computing is a centralizing force in that providers
realize economies-of-scale — especially with respect to human
capital and commodity hardware
• But is it also a decentralizing force?
Cloud computing as decentralizing force
• While the public cloud broadly is a centralizing force, inside the
enterprise, cloud computing acts as a decentralizing force
• The cloud presents disruptive price/performance that allows for
freedom from internal IT schedules and pricing — and from
legacy enterprise hardware providers (e.g. “blades and SANs”)
• The net is reduced time-to-market for enterprise developers —
which is especially important in emerging areas like mobile...
• ...but growth of the cloud in the enterprise has reduced level of
control — and can compromise its economic advantage
A heterogenous future
• Enterprises want to retain the economics and freedom that the
cloud represents, while reasserting the economics and control of
a centralized IT organization
• This points to a heterogeneous future: much enterprise compute
will remain on-premises — but the public cloud will remain critical,
driving both innovation and economics
• The centralization/decentralization oscillation will remain, but the
oscillations will not be in the technology itself, but rather in its
deployment: public or on-premises
• There are three key determinants for public v. on-premises:
• Economics: Rent vs. buy; OPEX vs. CAPEX
• Risk Management: Security/compliance — and also risk
factors associated with operator-as-threat
• Latency: The speed of light is a constant!
• These are all factors, but economics dominates: “private cloud”
efforts that do not deliver public cloud economics will fail!
The public/on-premises disconnect
• There are surprisingly few stacks that run both a multi-
tenant public cloud and are available as a software product
• AWS doesn’t (appear to) believe in an on-premises cloud
— and leading public clouds are not based on OpenStack
• For us at Joyent, both operating a public IaaS cloud and
shipping its orchestration software (SmartDataCenter) has
led to better engineering discipline and a superior artifact!
• Viz.: many of our architectural decisions came from the kiln
of unspeakable pain that is operating a public service
From heterogenous to hybrid?
• To make the leap from a heterogenous cloud to a hybrid
one, must have a common substrate that is run both on the
public cloud and on-premises
• Not enough to have mere “API compatibility”; for workloads
to truly straddle the cloud, gritty details like authentication/
authorization/accounting matter a great deal
• This implies not just technical hurdles but organizational
ones — and as a result, “true” hybrid cloud computing does
not feel close at hand...
So where is compute headed?
• As it always has, economics will chart the course
• Public versus on-premises will not be one decision, but
many — and many large enterprises will choose both
• The common substrate will be elastic infrastructure and
commodity hardware; it is not a choice between cloud
computing and legacy enterprise glop!
• We believe that there will be more uniﬁed providers that
make available both infrastructure-as-a-service and the
software to run infrastructure-as-a-service!