Juniper aims to achieve that simplification through designing network hardware and software in a way that allows network operations to be secure and automated.
Automation "is very near and dear to my heart", said Koley, who previously spent 10 years running Google's production network. Adding a layer of automation doesn't work very well; rather, "you have to design and engineer your system to be automated".
To that end, Juniper's Junos was the first network operating system to provide an API and fine-grained telemetry, he said. Large service providers, such as telcos and cable companies, need automation, and large cloud providers (a group that buys about 70% of their gear from Juniper) demand automation, so "automation is front and centre to what we are doing".
Furthermore, if you expect your network to scale by one or two orders of magnitude, automation becomes an existential issue because you won't be able to expand your network staff ten- or 100-fold to keep things running.
And it is important to realise that in very large networks there will always be at least one component — such as a switch or a router — that is broken at any given time, so you need software that can work around hardware problems.
One Juniper customer is feeding telemetry from its network into a machine learning system that can predict outages and automatically reroute traffic to avoid the ill-effects – with no human involvement.
Reliable networks are hugely beneficial, Koley observed. While at Google, he noticed that about 70% of network outages were due to human error, and they often occurred because the person concerned was tired or having a bad day. It's very difficult to reduce these errors, so the way around them is to greatly increase automation – and doing made Google's network more reliable.
But switching from human-based to automatic network management requires a change in mindset, Koley warned, and that makes it very hard to achieve 100% automation in one go. So it is best to start in a small way by automating one particular aspect and then going from there.
Another issue is that very few companies have engineering and operations teams of the size and skill levels of those at hyperscale businesses such as Amazon, Google and Microsoft. So Juniper is building products that allow smaller — though in Australian terms still relatively large — organisations to build networks that can be automated to a similar degree.
In addition, such organisations are saddled with legacy systems, so most of the things they can do add to the complexity rather than reducing it.
"You cannot wash complexity away. You have to engineer simplicity... That is our mission." said Koley.
In today's multi-cloud world, Juniper was working to provide seamless connectivity, orchestration across the whole fabric, end-to-end visibility, and security across boundaries, he claimed.
This would "solve complexity by engineering simplicity".