- Android Wear vs Apple Watch (Luke Wroblewski) — comparison of interactions and experiences.
- Eric Brewer on Kubernetes — interesting not only for insights into Google’s efforts around Kubernetes but for: There’s so much excitement we can hardly handle all the pull requests. I think we’re committing, based on the GitHub log, something like 40 per day right now, and the demand is higher than that. Each of those takes reviews and, of course, there’s a wide variety of quality on those. Some are easy to review and some are quite hard to review. It’s a success problem, and we’re happy to have it. We did scale up the team to try and improve its velocity, but also just improve our ability to interact with all of the open source world that legitimately wants to contribute and has a lot to contribute. I’m very excited that the velocity is here, but it’s moving so fast it’s hard to even know all the things that change day to day. Makes a welcome change from the code dumps that are some of Google’s other high-profile projects.
- We Don’t Sell Saddles Here — Stewart Butterfield, to his team, on product development and quality. Every word of this is true for every other product, too.
- What is Privacy Worth? (PDF) — When endowed with the $10 untrackable card, 60.0% of subjects claimed they would keep it; however, when endowed with the $12 trackable card only 33.3% of subjects claimed they would switch to the untrackable card. […] This research raises doubts about individuals’ abilities to rationally navigate issues of privacy. From choosing whether or not to join a grocery loyalty program, to posting embarrassing personal information on a public website, individuals constantly make privacy-relevant decisions which impact their well-being. The finding that non-normative factors powerfully influence individual privacy valuations may signal the appropriateness of policy interventions.
Comparing different orchestration tools.
Most software systems evolve over time. New features are added and old ones pruned. Fluctuating user demand means an efficient system must be able to quickly scale resources up and down. Demands for near zero-downtime require automatic fail-over to pre-provisioned back-up systems, normally in a separate data centre or region.
On top of this, organizations often have multiple such systems to run, or need to run occasional tasks such as data-mining that are separate from the main system, but require significant resources or talk to the existing system.
When using multiple resources, it is important to make sure they are efficiently used — not sitting idle — but can still cope with spikes in demand. Balancing cost-effectiveness against the ability to quickly scale is difficult task that can be approached in a variety of ways.
All of this means that the running of a non-trivial system is full of administrative tasks and challenges, the complexity of which should not be underestimated. It quickly becomes impossible to look after machines on an individual level; rather than patching and updating machines one-by-one they must be treated identically. When a machine develops a problem it should be destroyed and replaced, rather than nursed back to health.
Various software tools and solutions exist to help with these challenges. Let’s focus on orchestration tools, which help make all the pieces work together, working with the cluster to start containers on appropriate hosts and connect them together. Along the way, we’ll consider scaling and automatic failover, which are important features.
Elevate automation through orchestration.
As sysadmins we have been responsible for running applications for decades. We have done everything to meet demanding SLAs including “automating all the things” and even trading sleep cycles to recuse applications from production fires. While we have earned many battle scars and can step back and admire fully automated deployment pipelines, it feels like there has always been something missing. Our infrastructure still feels like an accident waiting to happen and somehow, no matter how much we manage to automate, the expense of infrastructure continues to increase.
The root of this feeling comes from the fact that many of our tools don’t provide the proper insight into what’s really going on and require us to reverse engineer applications in order to effectively monitor them and recover from failures. Today many people bolt on monitoring solutions that attempt to probe applications from the outside and report “health” status to a centralized monitoring system, which seems to be riddled with false alarms or a list of alarms that are not worth looking into because there is no clear path to resolution.
What makes this worse is how we typically handle common failure scenarios such as node failures. Today many of us are forced to statically assign applications to machines and manage resource allocations on a spreadsheet. It’s very common to assign a single application to a VM to avoid dependency conflicts and ensure proper resource allocations. Many of the tools in our tool belt have be optimized for this pattern and the results are less than optimal. Sure this is better than doing it manually, but current methods are resulting in low resource utilization, which means our EC2 bills continue to increase — because the more you automate, the more things people want to do.
How do we reverse course on this situation? Read more…
Start exploring Kubernetes with minimal effort.
I had not looked at Kubernetes in over a month. It is a fast paced project so it is hard to keep up. If you have not looked at Kubernetes, it is roughly a cluster manager for containers. It takes a set of Docker hosts under management and schedules groups of containers in them. Kubernetes was open sourced by Google around June last year to bring all the Google knowledge of working with containers to us, a.k.a The people :) There are a lot of container schedulers or orchestrators if you wish out there, Citadel, Docker Swarm, Mesos with the Marathon framework, Cloud Foundry lattice etc. The Docker ecosystem is booming and our heads are spinning.
What I find very interesting with Kubernetes is the concept of replication controllers. Not only can you schedule groups of colocated containers together in a cluster, but you can also define replica sets. Say you have a container you want to scale up or down, you can define a replica controller and use it to resize the number of containers running. It is great for scaling when the load dictates it, but it is also great when you want to replace a container with a new image. Kubernetes also exposes a concept of services basically a way to expose a container application to all the hosts in your cluster as if it were running locally. Think the ambassador pattern of the early Docker days but on steroids.