Skip to content

Commit 1028fa1

Browse files
committed
Add notes on roche
1 parent 522c9cd commit 1028fa1

2 files changed

Lines changed: 57 additions & 0 deletions

File tree

docs/edgecase-2025/0-keynote.md

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
# Edgecase
Lines changed: 56 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,56 @@
1+
# Navigating the Healthcare Multiverse
2+
3+
> How Roche United Cloud and Edge with a Platform Engineering Saga
4+
5+
## What is a lab?
6+
7+
When you give a blood sample or anything to analyze, the sample is sent through a pipeline
8+
of machines. The reality in the lab is fragmented, it's not an orchestrated pipeline. Machines have
9+
different form factors, cables are dangling, it doesn't nicely fit together.
10+
11+
### The problem
12+
13+
How can we fix the chaos at the edge? How do we provide reliable computing?
14+
15+
> Diversivied assets are great for your portfolio, not for your finances.
16+
17+
The stack must be easy enough for field service representatives to install them and flexible enough
18+
for developers to make their lives easier.
19+
20+
## The idea
21+
22+
Roche instruments are installed in a lab, Roche ships edge compute infrastructure that can talk to the
23+
lab machines.
24+
25+
### Phase 1 - Assembling the toolkit
26+
27+
You want to manage applications around the world consistently. Back in the days, there was really only a single
28+
solution, Rancher. Here's the challenge: you don't get inbound connectivity to hospitals, you're lucky if you
29+
get decent outbound connectivity.
30+
31+
As an operating system, Roche picked Debian based flavors. It's easy and security, no need to make it more complex.
32+
33+
How do you scale GitOps in this scenario? They used a templating engine based on jinja2 to render configuration per
34+
distribution. They have an entire KubeCon talk about this, but I'm not going deeper into this.
35+
36+
The next challenge is data resilience. How do you make sure you operate while keeping patient data local?
37+
38+
Here comes their 3 tier architecture. Global infrastructure manages regional control planes, regional control
39+
planes manage edge infrastructure.
40+
41+
### The evolution
42+
43+
The first evolution was building a fleet manager. Rancher was not really handling edge behaviour where laptops
44+
are being closed and cables are being disconnected. So Roche started building out an in-house fleet management console.
45+
46+
The single IP address was becoming a bottleneck. To solve this, they leveraged Cilium mTLS tunnel to connect
47+
their edge compute to compute in the cloud. From the location in the cloud, they can connect to anywhere.
48+
49+
As an operating system, they moved to Talos to reduce the bundle size they were shipping to the customers.
50+
51+
## Links
52+
53+
- <https://github.com/danacr/kubernetes-the-fun-way>
54+
- <https://developers.cloudflare.com/reference-architecture/diagrams/network/bring-your-own-ip-space-to-cloudflare/>
55+
- <https://docs.cilium.io/en/latest/network/servicemesh/mutual-authentication/mutual-authentication/>
56+
- <https://github.com/Roche/foxops>

0 commit comments

Comments
 (0)