This video is presented by Yuval Bachar, the same person posted this announcement on the Facebook Blog.
Here are some quick notes I took after watching this video.
My quick notes of this video
Wedge only provides 16 40G ports instead of 32 ports the chip can support. This is to make Wedge a building block of "6-pack".
6-pack is equal to 12 Wedges stacked together. (I noticed this point from my earlier post).
Agility: 3 months for building Wedge, and 4.5 months for "6-pack".
We (Facebook) wrote all the BGPs.
Everything is on Linux.
6-pack is 100Gbps ready when silicons are stabilized.
6-pack is switching-ASIC agnostic.
In 6-pack, no special buses. Everything is Ethernet (internally).
With 6-pack, every element (line cards, fabric cards) is a server. We (Facebook) can add new features on these "servers" very quick.
At 13:29, we can see the photo of the back side of 6-pack.
Every element of 6-pack controls its own 1+1 cooling fans.
Gigabit Ethernet is used as IPC (Inter-Process Communication?) among the elements of a 6-pack.
It can grow to higher radix version when needed, such as 12-pack or 24-pack.
Our (Facebook) goal is to share our knowledge to the world to enable people to build better data center.
Facebook is not going to sell it. We cannot order "6-pack" from Facebook.
One more thing...
A fabric card of 6-pack is equivalent to two 32-internal-port-only Wedges bounded together. Two fabric cards are equal to 4 Wedges. Therefore, a single 6-pack is equal to 8+4=12 Wedges stacked together.
High-level “6-pack” block diagram and the internal network data path topology. Copied from Facebook's original post. |
In my earlier post, I used Cisco's Nexus 9332PQ box as an example of individual switching element instead of Wedge. The only reason I picked Cisco's box is: 9332PQ is another 32x40Gbps non-blocking full line rate Layer 3 switch just like Wedge.
No comments:
Post a Comment
Tip: you can also anonymously comment here.