AWS re:Invent 2022 - Dive deep on AWS networking infrastructure (NET402)

[@vumilemncwango7962]

Amazing for any company to do what AWS has done in networking

[@moe85moe85]

Really great talk, look behind the scenes and the complex engineering of keeping the building blocks simple and scalable

[@sats_nak]

loved teh deep dive here. More next yaer pls!

[@kellymoses8566]

This was incredibly interesting.

[@awssupport]

I'm so pleased to hear that you found this video interesting, Kelly! 👍 ^NR

[@stephenc123]

Great to hear!

[@Soulseeologia]

Thanks this helps me in AWS my teacher’s a senior they’ll be a principle one day!

[@KenSherman]

Can I also consider this as an illustration of AWS's anatomy of their backbone?

[@RandyRam9]

Really, really great presentation! 👏

[@awssupport]

Glad to hear you enjoyed this, Randy! ☁️ ^RN

[@kiranjbm]

Really nice presentation

[@awssupport]

Thanks! Best wishes as you continue to dive deep. 🤿 ^JM

[@dus10dnd]

WE NEED STRETCHY SUBNETS! Please! Let's have them stretch across AZs!

[@venkat290485]

Yes, We need this. Its Possible on Huawei Cloud.

[@timothyhendricks3004]

That would defeat the purpose of availability zones.

[@dus10dnd]

@@timothyhendricks3004 No it wouldn’t. It works perfectly in both Azure and GCP. And let me elaborate... the point of availability zones is not about networks. It is about having workloads that can withstand a datacenter outage. Stretching a subnet does absolutely nothing to diminish that, but it does enhance the experience in a number of ways. First, you will deploy multiple workloads, but in separate AZs... a datacenter failure isn't the only reason that you deploy mulitple workloads... an individual resource can fail for many reasons and the current AZ approach is tightly coupled. To really make use of an AZ, you want to duplicate your entire stack and ensure that it routes within the AZ to the other resources to which it is dependent. Stretching a subnet would still allow for this (would also require NLBs that stretch AZs with a single endpoint), but it would let those redundant workloads work better with a failed component in the stack by automatically allowing it to participate with the remainder of the stack. In addition, subnetting sucks... I know... I am really good at it. Creating a separate subnet for each AZ consumes many IPs... the network address, the broadcast address... the gateway IP... other reserved IPs. And other good practices that make use of subnets... now you need 2, 3, 4... Nx subnets. No thanks. It is a bad design. AWS needs to go back to the drawing board of many aspects of their architecture and rev it. They've been around for far to long to keep living on their incremental improvements. Microsoft completely revved their API and datacenter architecture in 2014 and it made a massive difference.

[@brijeshmondal_]

Great content wonderful presentation🤘

[@AWSEventsChannel]

😀 🙌

[@dcastelltort]

Very interesting, thank you

[@awssupport]

Hi Damien! 👋 We're happy to hear that you enjoyed it. 😀 ^AK

[@declanmcardle]

@1:00:37 - oh look, it's eu-west-1 at night 🙂

[@gormicz]

Awesome

[@declanmcardle]

@44:00 Spot the Simpsons reference 🙂

[@no-bc4kc]

Tenets??🤔🤔🤔 MAKES ONE WONDER

[@GP-ez5ms]

First 15 minutes were complete marketing BS. Couldn’t go any further

[@igp2bgp]

LOL, that BS came from cisco fellow, so worth listen :)