Hey, network giants: Facebook swigs from an 'open' 6-PACK of tech

Facebook has taken another step towards kicking the traditional switch vendors out of its network, setting their chassis-based switches in the cross-hairs. The Social NetworkTM has created a chassis and fabric for the Wedge switches it let loose on its data centres last year. Announced at today's Facebook Networking@Scale …

[Voland's right hand]

This diagram makes very little sense

It is probably me being thick today, but this diagram does not make a lot of sense :)

[Tom Samplonius]

Re: This diagram makes very little sense

"It is probably me being thick today, but this diagram does not make a lot of sense :)"

It is only the control plane for the switch. So the "ASIC boxes" are hiding a lot of detail. This diagram just shows how the microservers and various control bits talk to each other.

[P. Lee]

FB does something worthwhile?

This looks as though it will be an interesting situation.

[Anonymous Coward]

Re: FB does something worthwhile?

Outside its core business Facebook does a few worthwhile things.

It's either acute irony or just good business that the proprietary hardware vendors are under pressure from open information initiatives in software and compute being incubated by hyperspace proprietary information hoarders.

Having cut - throat commodity competition in every layer of the stack except your own is good for business.

[OllyL]

For those of us unable to convince a white-box vendor to build one for us, will these be available to the unwashed masses at any point to buy, or will it just be something for us to marvel at from afar?

[Tom Samplonius]

"...will these be available to the unwashed masses at any point to buy, or will it just be something for us to marvel at from afar?"

Well, the design has been contributed to the Open Compute Project, so probably all white box makers will be shipping these.

So try http://www.quantaqct.com and http://www.pluribusnetworks.com/

[DataCenter_Geek]

Unfair comparison

The comparison to the Cisco 9516(which can support 576 x 40g ports ) is unwarranted as the 9516 is a 16 line card chassis and 6-pack is a 4 line card chassis. It can however be compared to the Nexus 9504 which allows for 144 x 40G ports and 6-pack which will have 128 x 40g ports which is still lacking but close.

Any idea what chipset is being used for the line card Wedge and what processor is powering those supervisors.

[Tom Samplonius]

Re: Unfair comparison

"Any idea what chipset is being used for the line card Wedge and what processor is powering those supervisors."

First of all, it is not a Cisco switch, so there are no supervisors. Each line card has a microserver, based on the Group Hug standard. The Group Hug standard is basically a pluggable single board PC. So normally, it would be a low-power Intel or AMD. But it could be ARM. Or something else, as Group Hug is a standard motherboard interface, so the switch is CPU agnostic. Typically, the microserver would run Linux with an OpenFlow service. The microserver's would be the closest equivalent to a supervisor.

Cisco cycles between including the fabric on the supervisor, or making the fabric a separate card (during the long life of the Catalyst 6500, Cisco tried both). The 6-pack has "dumb" fabric cards, that are under control of the microservers.

[Crazy Operations Guy]

As bad as Cisco for the naming

I hate it when the models numbers / names of devices imply a certain number of slots for network cards in the things, but then you find out that two are being used for the management blades.

[Tom Samplonius]

Re: As bad as Cisco for the naming

"...management blades."

Strictly speaking, those are the fabric blades. The management is on the line cards.

[Alistair]

I need to ......

Cobble one or two of these together for the next LAN party...

Yup,

DOOM at 40Gbs!!!!

Having to mangle several large data management/manipulation environments into our Not Yet Ready for 40Gbs backbone environment, these things make interesting reading.