OpenStack has come a long way since 2010. What started as a collaboration on compute and storage between NASA and Rackspace has changed dramatically and grown into a large, successful open source project that meets the needs of thousands of organizations. But OpenStack hasn’t evolved in a vacuum over the past seven years: the technology landscape around it has been changing as well. Join VMware’s chief OpenStack architect and longtime community member Mark Voelker for a look at the new technology landscape around OpenStack, how we got here, and where we might go next. We’ll discuss how what started as an IaaS platform ending up being a winning platform for Network Functions Virtualization and telco applications, how OpenStack came to be selected as a common underpinning for container orchestration systems like Kubernetes, how OpenStack governance influenced other open source communities, and how OpenStack changed the way companies looked at Open Source. We’ll consider the role IaaS might play in a future that includes options like functions-as-a-service, containers, and the internet of things. We’ll consider OpenStack as a common foundation for a variety of new technologies, and discuss OpenStack’s lasting impact in the cloud ecosystem. We’ll also discuss how OpenStack is changing and adapting to shifts in the technology landscape, both as an open source community and in terms of product offerings. Learn about new interoperability programs targeted at use cases that didn’t exist seven years ago, and new initiatives from the OpenStack technical community and Foundation.
2. The Year Was 2010…
2
A newly announced “cloud operating system” from Rackspace and NASA immediately stood out.
(things moved pretty quickly from there…)
3. Why was this the right moment for OpenStack?
3
There were (and are!) lots of other great clouds….
And where is OpenStack going next?
4. • Chief OpenStack Architect @ VMware since 2014
• OpenStack Community: Interop WG Co-chair, OpenStack ATC & AUC, Former Puppet-OpenStack core
dev, Triangle OpenStack Meetup founder, OpenStack Foundation Member #54
• OpenStack community member since 2011
• Unfortunately has no super powers at all other than ability to consume large quantities of donuts
“A computer nerd….is somebody who uses a computer in order to use a computer.”
–Douglas Adams
Mark T. Voelker
5. 5
In 2010 we were already aware of how
powerful this new “cloud” operating
model could be….
6. 6
Also: the world already had a lot of servers, and was well into the journey to virtualize them...
“…around 50 percent of x86 architecture server workloads [will be virtualized] by the end of 2012 [up
from 16% in 2009]” --Gartner, 2009
“…by year-end 2010, enterprises with 100-999 employees will have a higher penetration of virtual
machines deployed than the Global 500.” --Gartner, 2009
7. 7
Virtualization (& the networking, storage, etc advancements that came with it)
had given us a lot of power.
If we could build on those investments and create an
even more powerful cloud…that made a lot of sense!
8. Maybe even more important: the game was now all about going fast.
8
More than just new tools: tearing down silos.
Infrastructure at the speed of code.
9. 9
CapEx dollars could stretch further than ever!
Everyone wanted a crowd of talented
developers. They’re available all over the world
and there’s lots of them!
Developers are increasingly training with a
relatively newly-accepted weapon in
infrastructure: open source.
10. 10
But it turns out major technology
transitions can be tough for
organizational reasons.
• Rent vs buy? Actually a big
change for budgeting models.
• 0% of CFO’s polled in 2010
thought giving everyone a credit
card and letting them file
expense reports for compute
was a more awesome idea than
an IT budget.
• Changing runbooks, tools,
monitoring, lifecycle, and
people’s technical skills adds
up.
• Dismantling silos sometimes
leads to friction even within
strong organizations.
• Devs tend to specialize…just
how big is that pool of
specialized cloud-and-
distributed-systems developers?
• “Software is eating the world”
may be true, but it’s not a
business plan.
11. 11
In major technology transitions, winning is rarely about using something new.
Success is about making the right transition to a new model.
So, it’s all
about how you
get there.
12. 12
And we picked up some awesome tools in the early 2010’s…
• SDN
• ONF founded 2011
• VXLAN & NVGRE drafts published early 2012
• Big Data
• Hadoop widens adoption, Cloudera founded in 2007
and HortonWorks in 2011
• Spark released in 2014
• Distributed systems
• Zookeeper 1.0 in 1999
• RAFT paper published in 2014
• Cloud
• EC2 exits beta in 2008
• Azure commercially available in 2010
• CloudFoundry announced by VMware in 2011
• All this changed the datacenter:
• East-West traffic eclipses North-South (return of CLOS
and leaf-spine networks)
• Multi-tenancy
• Much higher density, much more automation
13. 13
OpenStack gave us a way to incorporate both
new and battle-tested tech together.
We could use familiar things and processes, as
well as new tools and interfaces.
OpenStack’s integration
engine helped us mortals
make the transition to cloud.
14. 14
Truth time: why does VMware care?
We’re a software infrastructure company.
Catching transitions is pretty much what we do.
15. 15
For us, OpenStack turned out to be hugely
complimentary!
• Virtualization?
• #1 hypervisor in the industry
• Storage?
• Huge ecosystem of storage
providers with an HCL and loads of
testing
• VMware vSAN
• Network?
• NSX was one of earliest success
stories for OpenStack networking.
• Just want the basics? DVS drivers
too!
• Management
• vROps, vRA, LogInsight, vRNI
• Billing
• vRealize Business
• Ecosystem of third party tools
• Massive number of existing workloads
Remember: it’s about transitions.
16. 16
It wasn’t hard to see the potential power here.
It also helped us discover
all kinds of new things that
we didn’t think we could do
before.
OpenStack helped us
integrate it all.
17. 17
Let’s add that all up. What’s good:
Fast
Developers
Smooth transitions
Open Source
Open source cloud helps devs go fast & organizations make smooth transitions.
(There’s your hockey stick growth curve for OpenStack, folks.)
19. 19
This year I’ve logged over 170,000
miles traveling to four continents and
talking to hundreds of people working
with OpenStack.
I see two use cases for OpenStack
picking up more steam than any
others:
• IaaS supporting container
orchestration systems
• Telco/NFV applications
Important note: other use cases are
also appearing, and “traditional” ones
are growing too!
20. 20
Changes are being made
throughout the OpenStack
ecosystem as a direct result of
these.
THIS.
IS.
EXCITING.
21. OpenStack is now much more
mature and widely adopted than
it was years ago.
It’s also incredibly flexible and
powerful.
So what I consider to be a
natural (and encouraging!) thing
is now happening…
21
OpenStack is specializing
to address new use cases.
(without leaving existing ones behind)
22. 22
“Given time, every sufficiently good and flexible platform
trends toward specialization.”
Consider Linux: it’s in your servers, your phone, your
seatback entertainment unit, your nuclear submarine,
your router, and your factory robots.
(but different spins for different usecases)
23. 23
If you’re going to specialize, don’t do it halfway.
(You want a new armored suit, not just a different helmet.)
That means you need the whole ecosystem involved.
24. 24
What kinds of changes? Just a few examples:
Things changing in code:
• Extended EPA support
• VLAN-aware VM’s
• Cells v2 & placement API
• PCI weighter
• Improved SR-IOV support (such as egress QoS rules
for SR-IOV)
• Improvements to Ironic and Magnum
Foundation initiatives:
• Focus on composable infrastructure and improved
support for standalone services
• Vertical interoperability programs
• OpenDev conference
• New and growing SIGs and cross-participation in
adjacent communities like ONAP, OPNFV,
Kubernetes
Vendor ecosystem:
• Deployment options from HA to compact
• Migration of existing workloads
• N+2 upgrades and longer support cycles
• Hitless upgrades
25. How We Do Upgrades: Blue-Green Upgrade Pattern
25
Load Balancer
• Zero dataplane downtime throughout.
• Allows new control plane to be tested before going live
• Very fast rollback
• Allows for root causing of problems since both planes can be kept in event of failure
• Skipping releases? No problem. Doesn’t depend on n-1 or n-2 compatibility in control plane components.
• Eases addition of new components/decomposition since green plane is “just a new deploy”.
• Leverages existing deployment codepath == less complexity.
Mitaka Control Plane Ocata Control Plane
26. 26
We’ve still got work to do.
• Service chaining
• More scalability (up as
well as down)
• Maturation of newer
features
• Easier composition of
infrastructure—run just
the services you want
• Vertical interop programs
• Bugs sometimes happen
• Docs sometimes don’t
happen
• Building the use cases,
sharing what works
27. 27
We can be even more powerful
with a few good friends!
• Many OpenStack
community members work
in other communities too.
• SIGs help bring our
communities together.
• OPNFV, Kubernetes have
developed/are developing
standards programs similar
to OpenStack Powered.
• Continued outreach will be
important for OpenStack
going forward.
28. 28
“Engineering is science
spike with art and stirred
with a dollar sign.”
Laser focused on cutting
costs? Don’t.
• Focus on delivering
value instead.
• The CapEx fallacy
• Get the transitions right.
Worried about ecosystem
changes? Don’t be.
• A good ecosystem is one
where bets are made.
• Not all succeed.
• It’s ok if some players
consolidate or exit.
• There are zero
successful tech
ecosystems where this
hasn’t happened.
29. 29
If we look even further out: more
changes coming in the
infrastructure space!
• SD-WAN?
• Unikernels?
• AI and machine learning?
• New security stuff?
And more of the same:
• Keep the core stable
• Keep the options compsoable
• Flexibility is key
OpenStack has come a long way since 2010. What started as a collaboration on compute and storage between NASA and Rackspace has changed dramatically and grown into a large, successful open source project that meets the needs of thousands of organizations. But OpenStack hasn’t evolved in a vacuum over the past seven years: the technology landscape around it has been changing as well. Join VMware’s chief OpenStack architect and longtime community member Mark Voelker for a look at the new technology landscape around OpenStack, how we got here, and where we might go next. We’ll discuss how what started as an IaaS platform ending up being a winning platform for Network Functions Virtualization and telco applications, how OpenStack came to be selected as a common underpinning for container orchestration systems like Kubernetes, how OpenStack governance influenced other open source communities, and how OpenStack changed the way companies looked at Open Source. We’ll consider the role IaaS might play in a future that includes options like functions-as-a-service, containers, and the internet of things. We’ll consider OpenStack as a common foundation for a variety of new technologies, and discuss OpenStack’s lasting impact in the cloud ecosystem. We’ll also discuss how OpenStack is changing and adapting to shifts in the technology landscape, both as an open source community and in terms of product offerings. Learn about new interoperability programs targeted at use cases that didn’t exist seven years ago, and new initiatives from the OpenStack technical community and Foundation.
But if this was just about servers, this whole movement would have been really short. The average lifespan of a production server was 3-5 years.
You don’t get fired for buying $server/$net/$storage, you get fired for buying it and then chucking it out the window to use the cloud a year later. And before anyone says “sunk costs fallacy”, ask yourself if you’d buy a new car every time a faster/roomier/more fuel efficient model comes out….and then ask yourself if you were a trucking company if you’d replace your entire fleet.
If you have fast, you can make your own good and cheap.
I want fast, fast, and fast with a side of fast.
Developers: who wouldn’t rather invest in people than thankless boxes? That’s rad! People can learn new tricks quickly; boxes are pretty static.
New developers increasingly turned to open source as they learned. And that was now accepted in the datacenter too: everyone had Linux, most had Apache, many had MySQL. Nobody ever got fired for building on the LAMP stack.
Distributed systems are still hard (hands up if you’ve ever written a perfect PAXOS implementation! Nobody? How about if you actually have a thorough understanding of it?)
Open Source is great for creating general-purpose systems.
Try a lot of things with ideas from potentially LOTS of different backgrounds.
Once there’s consensus: parallel development, across the world. It’s stone soup with 0’s and 1’s!
A few players leaving the game doesn’t dent progress.
Pooling risk looks super-good right after a financial crisis by the way.
If you want to capitalize directly on the open source thing…that can be tough.
”Make it boring” certainly works, but there won’t be a lot of players.
But if you’re capitalizing not on the thing, but on something above or below it…
On that last point: I was on the customer advisory council for a cloud provider in the early 2010’s. What did I tell them I wanted? Totally unsupported cheap infrastructure and I didn’t care (much) if it sometimes broke. Why? I wanted to run temporary experiments and I had apps that were distributed, not centralized.
Proud moment: VMW is catching these transitions well. You already had a datacenter running ESXi, so wrapping a new API and tenancy on it made sense. You needed the network too though—hence, NSX. Then you needed more horizontal infra: NSX-T + Multi-VC. Then you needed containers: VIO-K and PKS.
And by the way: we also had a global financial crisis in 2008 that had people thinking a lot about being more agile while also being able to stretch their existing resources further.
We didn’t have to throw out all our old training and processes. We could use that storage array we just bought, but add a new SDN platform to make our network more powerful. Our hypervisor guy could troubleshoot just like before, but nobody had to file tickets to get a VM anymore. We didn’t have to replace the AD server or map it to some new system.
E-commerce, financial services, data processing, dev/test environments, etc etc etc.
Serverless (on the IaaS? On the CaaS?), Machine Learning (will unlock the next wave of distributed systems), Edge Computing (remember cycles? Time to swing back to the edge…and don’t forget SD-WAN), IoT
Keep your core solid, specialize carefully for emerging use cases.
Do this well, and we’ll still be doing Summits in 2027.
Note: this slide is animated to show the logical flow