couchbase containers with bare metal performance – couchbase live new york 2015
TRANSCRIPT
Couchbase Containers withBare Metal Performance
Bryan Cantrill
@bcantrill
Elastic infrastructure, circa Dot Com Boom
• In the late 1990s, the only way to meaningfully scale a database was up — and the physical infrastructure had to scale with it
• This was excruciatingly expensive — and became a non-starter in the post-apocalyptic nuclear winter of the early 2000s...
Elastic infrastructure, circa Dot Com Bust
• The rise of rack-and-stack commodity servers brought with it new distributed software architectures like memcached that were designed to scale across many machines
• The rise of these architectures afforded new operational possibilities: if the computer itself is a commodity, why buy it at all? Why not rent from someone who runs it cheaper and better?
• From the perspective of compute providers, economies of scale could only be realized if hardware is shared across tenants...
• Multi-tenancy demands virtualization, but where in the stack to virtualize?
Hardware-level virtualization?
• The historical answer to virtualization — since the 1960s — has been to virtualize the hardware:• A virtual machine is presented upon which each tenant runs an
operating system that they choose (and must manage)
• There are as many operating systems on a machine as tenants!
• Can run entire legacy stacks unmodified...
• ...but operating systems are heavy and don’t play well with others with respect to resources like DRAM, CPU, I/O devices, etc.
• Hardware-level virtualization limits tenancy and performance!
Platform-level virtualization?
• Virtualizing at the application platform layer addresses the tenancy challenges of hardware virtualization, and presents a much more nimble (& developer friendly!) abstraction...
• ...but at the cost of dictating abstraction to the developer
• This is the “Google App Engine” problem: developers are in a straightjacket where toy programs are easy — but sophisticated applications are impossible
• Virtualizing at the application platform layer poses many other challenges with respect to security, containment, etc.
OS-level virtualization?
• Virtualizing at the operating system level began with the crude filesystem virtualization of chroot in Seventh Edition Unix
• chroot originated with Bill Joy, but specifics are blurry; according to Kirk McKusick, via Poul-Henning Kamp and Robert Watson:
OS-level virtualization
• Seeking to provide a security mechanism, FreeBSD extended chroot into jails:
• To provide workload consolidation, Sun introduced complete operating system virtualization with zones (née Project Kevlar)
OS-level virtualization
OS-level virtualization!
• A single operating system (i.e. a single kernel) allows for efficient use of hardware resources, maximizing tenancy and performance
• Disjoint instances, known as containers, are securely isolated and virtualized by the operating system
• Gives tenants what appears to be a virtual machine (albeit a very fast one) on which to run higher-level software
• Containers combine the developer ease of platform-level virtualization with the generality of hardware-level virtualization!
OS-level virtualization at Joyent
• Joyent runs OS containers in the cloud via SmartOS — and we have run containers in multi-tenant production since ~2006
• Adding support for hardware-based virtualization circa 2011 strengthened our resolve with respect to OS-based virtualization
•We emphasized their operational characteristics — performance, elasticity, tenancy — and for many years, we were a lone voice...
Containers as PaaS foundation?
• Some saw the power of OS containers to facilitate up-stack platform-as-a-service abstractions
• For example, dotCloud — a platform-as-a-service provider — built their PaaS on OS containers
• Struggling as a PaaS, dotCloud pivoted — and open sourced their container-based orchestration layer...
...and Docker was born
Docker revolution
• Docker has used the rapid provisioning + shared underlying filesystem of containers to allow developers to think operationally
• Developers can encode deployment procedures via an image
• Images can be reliably and reproducibly deployed as a container
• Images can be quickly deployed — and re-deployed
• Docker will do to apt what apt did to tar
Broader container revolution
• The Docker model has pointed to the future of containers
• Docker’s challenges today are largely operational: network virtualization, persistence, security, etc.
• Security concerns are not due to Docker per se, but rather to the architectural limitations of the Linux “container” substrate
• For multi-tenancy, state-of-the-art for Docker containers is to run in hardware virtual machines (!!)
• Deploying OS containers in hardware virtual machines negates their economic advantage!
Container-native infrastructure?
• SmartOS has been container-native since its inception — and running in multi-tenant, internet-facing production for many years
• Can we achieve an ideal world that combines the development model of Docker with the container-native model of SmartOS?
• This would be the best of all worlds: agility of Docker coupled with production-proven security and on-the-metal performance of SmartOS containers
• But there are some obvious obstacles...
Docker + SmartOS: Linux binaries?
• First (obvious) problem: while it has been designed to be cross-platform, Docker is Linux-centric — and the encyclopedia of Docker images will likely forever remain Linux binaries
• SmartOS is Unix — but it isn’t Linux…
• Fortunately, Linux itself is really “just” the kernel — which only has one interface: the system call table
•We resurrected (and finished) a Sun technology for Linux system call emulation, LX-branded zones, the technical details of which are beyond the scope of this presentation...
LX-branded zones: tl;dr
Docker + SmartOS: Provisioning?
•With the binary problem being tackled, focus turned to the mechanics of integrating Docker with SmartOS provisioning
• Provisioning a SmartOS zone operates via the global zone that represents the control plane of the machine
• docker is a single binary that functions as both client and server — and with too much surface area to run in the global zone, especially for a public cloud
• docker has also embedded Go- and Linux-isms that we did not want in the global zone; we needed to find a different approach...
Aside: The power of an interface
Aside: The power of an interface
Aside: The power of an interface
Aside: The power of an interface
Docker Remote API
•While docker is a single binary that can run on the client or the server, it does not run in both at once…
• docker (the client) communicates with docker (the server) via the Docker Remote API
• The Docker Remote API is expressive, modern and robust (i.e. versioned), allowing for docker to communicate with Docker backends that aren’t docker
• The clear approach was therefore to implement a Docker Remote API endpoint for SmartDataCenter, our (open source!) orchestration software for SmartOS
Triton: Docker + SmartOS/SmartDataCenter
• In March, we launched Triton, which combines SmartOS and SmartDataCenter with our Docker Remote API endpoint
•With Triton, the notion of a Docker host is virtualized: to the Docker client, the datacenter is a large Docker host
• One never allocates VMs with Triton; all Triton containers are run directly on-the-metal
• All of the components to Triton are open source: you can download and install SmartDataCenter and run it yourself
• Triton is currently general available on the Joyent Public Cloud!
Container landscape
• It is becoming broadly clear that containers are the future of application development and (especially) deployment
• But the upstack ramifications are entirely unclear — there are many rival frameworks for service discovery, deployment, etc.
• The rival frameworks are all open source:
• Unlikely to be winner-take-all
• Productive mutation is not just possible but highly likely
• Triton takes a deliberately modular approach: the container as general-purpose foundation, not prescriptive framework
Containers and Couchbase
• Couchbase is particularly appropriate for containers: its scale-out architecture demands elastic infrastructure — and its use cases demand on-the-metal performance
• But hardware-virtualized Docker hosts undermine the efficacy of containers — and force an allocation-oriented disposition instead of allowing a consumption-oriented one
• The Triton model represents Couchbase containers without compromise: like Couchbase itself, the infrastructure can grow as needed — while still delivering bare-metal performance!
Future of containers
• For nearly a decade, we have believed that OS-virtualized containers represent the future of computing — and with the rise of microservices + Docker, this is no longer controversial
• But to achieve the full promise of containers, they must run directly on-the-metal — multi-tenant security is a constraint!
• The virtual machine is a vestigial abstraction; we must reject container-based infrastructure that implicitly assumes it
• Triton represents our belief that containers needn’t compromise: multi-tenant security, operational elasticity and on-the-metal performance!