This is much more used, it's based on FAIRSeq Transformer: https://github.com/modernmt/modernmt.
Plus, it's a publicly funded project
670 karma · joined April 15, 2014
Day to day activities can be proficiently carried out through laptops, while heavy crunching and 24/7 operations must be carried out through heavily cooled rigs.
So, what's the place for a fanless, low powered mini pc?
Hardware is super cheap:
-A 40 slots rack, with gigabit fiber, dual power and a handful of public IP addresses, costs on average 10000€/y.
-A reconditioned server on eBay with 16 cores and 96GB of RAM costs 500€ (never seen them break in 3 years).
-A brand new Dell Poweredge with AMD EPYC 32 core and 64GB of RAM will cost 3000€.
-Storage is super cheap: 500GB of SSD costs 80€ (consumer stuff is super fine as long as you wisely plan between redundancy and careful load) and rotational disks are even cheaper. Never seen a rotational disk break.
Once bought, all of this is yours forever, not for a single month. You can pack very remarkable densities in a rack and have MUCH more infrastructure estate at disposal than you would ever afford on AWS.
The flip side of the coin is that you need operation expertise. If it's always you, then ok (although you won't be doing much more than babysitting the datacenter). Otherwise, if you need to hire a dedicated person, people is the most expensive resource yet and that should definitely be added to the cost of operations.
Containers were never meant to be "secure" and Kubernetes has nothing to do with security in general. Application level containerization solves 3 problems: 1- repeatable builds 2- software distribution 3- infrastructure mutability
Docker solves brilliantly all these problems: we can build software in controlled environments that we call images and the product is an image itself, that we can share by means of software catalogs called registries and that we can deploy the same way we can play CDs in a computer, without modifying the host system. Security was never meant to be a concern, it was never part of the plan, at least no more than running a process in an operating system. And containment systems are just a smart way to launch a process, that's it. They can't replace security functionalities beyond what is already provided by host OS.
Is this a fault? Would you blame a hammer for not being able to unscrew a cork?
NanoVMs CEO spends some time also making fun of "stripped down systems", accusing them of recklessly "ripping out components of a system that simply don't belong in the virtualized world of 2020", namely users, SSH logins and desktops. Again, there is a big misunderstanding here. Linux processes expect some facilities to be present at runtime, namely system calls to retrieve Process IDs and user IDs. But these facilities were merely retrofitted in containment systems to avoid processes crash upon syscall invocations. As all retrofitted functionalities, their implementation is barely beyond the dummy level. Therefore, we have PID 1 everywhere, no SSH logins and dummy users.
Now, Kubernetes. K8s is just a scheduler, an orchestrator. It has nothing to do with security, it is mainly concerned with resource optimal usage and container placement. And it is a fundamental component in microservice world, otherwise we would be paralized by operations at each deployment. "You spin up a linux box and call it a day - done". Seriously? Is this really the whole understanding of the world of operations? I've been in ops for 12 years and I'd never want to go back to the old bad days of 2008, where most of us managed an handful of servers and services and, yet, datacenter ops were at one with deployment ops. That was the very world of manual and error prone ops. Orchestration has been a bliss but requires a unit of scheduling. A container, in this case. But it may be a VM as well.
So, the saddest thing here is that NanoVMs could be a perfect replacement for containers while retaining the benefits of orchestration (as in Kubevirt, for example), but the CEO prefers to throw away the dirty water along with the baby for marketing hype.
"At one particular ‘top’ tech company the process is that when a candidate goes through an interview he or she has a packet compiled about their interview performance. The packet then goes to a committee whose job it is to impartially review the packet to make a hiring decision. At one point a particular committee got so critical that they rejected every packet for several months. When HR caught wind of this they decided to set up a test. They sent the committee a new round of packets and once again the committee rejected them all. HR then called them all into a meeting and explained that they packets they had just reviewed were in fact the hiring committee member’s own packets from when they interviewed for that company. They had unknowingly rejected themselves! How could anyone pass that bar?"
Here is a nickel kid: buy yourself a real computer -Dilbert
It's also graph analysis and ML models.
Isn't this actually illegal?