Has anyone here looked into Facebook FBoss project and have any experience in deploying the code in their depot? https://github.com/facebook/fboss
It looks pretty active.
I had heard something about them making their own switches but never read about it. This is really interesting considering we just had a discussion about smaller port count switches for top of rack applications.
I think its going to take a while for solution like this to build traction in the market. Everyone is too comfortable and reliant on the safety net that is vendor support. If companies do have a use-case for venturing out or building their own I think they would start with companies like Cumulus or Nuage and work from there.
I do welcome the day when the DC network switches are more like Linux servers, and if you do have super critical applications or OSs you can get vendor support similar with Red Hat.
And on this note look what popped up in my feed. Dell has their own OS now. This is going to be interesting!
http://networkingnerd.net/2016/01/20/the-tortoise-and-the-austin-hare/
Network Field Day 11 is going on right now and Dell is up today at 11am PST. This is going to be interesting!
Networking is so core that anyone that would have the IT staff to investigate this kinda stuff wouldn't be likely to implement it. It's kinda cliche but the only thing more expensive than a high-end networking vendor is the possibility of a network outage that comes with cheap home-made gear.
Personally I would like to play with them, but I have no intention of installing them in my data-center any time soon.
Quote from: dlots on January 21, 2016, 09:30:46 AM
Networking is so core that anyone that would have the IT staff to investigate this kinda stuff wouldn't be likely to implement it. It's kinda cliche but the only thing more expensive than a high-end networking vendor is the possibility of a network outage that comes with cheap home-made gear.
Personally I would like to play with them, but I have no intention of installing them in my data-center any time soon.
Traditional enterprise networks yeah I agree. But with the DC its different. They are built out very redundant. Racks are built with redundant ToRs and uplink to redundant spine switches. The larger you go the more pods you introduce that provide pod-redundancy. So if they are architecture correctly a failure in one pod or the whole pod will not affect other pods or production traffic.
To me its a shift in mentality from the tradition network thinking of unbox vendor-Xs switch update code, rack and configure to the same way a server is built. Unbox, install corporate tested OS and packages, verify functionality, then rack and build config. Obviously there is a ton of testing and verification before hand that what you are deploying on the switch functions as expected.
Even further server guys have already figured out how to automate and templatize this process with the various tools out there. These tools are slowly creeping into the network realm but would play a huge role in this scenario.
Quote from: that1guy15 on January 21, 2016, 10:33:09 AM
Even further server guys have already figured out how to automate and templatize this process with the various tools out there. These tools are slowly creeping into the network realm but would play a huge role in this scenario.
Truth... I ran into this the other day. The server guys use Jenkins+Ansible for automation and have been for years. Management now expects their network team to get it up and running in a few weeks. ;P