x

Google's Andromeda Strain Is Spreading

Google is opening up its "Andromeda" SDN architecture for customers using its Connect Engine system to develop cloud services. Google (Nasdaq: GOOG) revealed in a blog post that developers using the "us-central1-b" and "europe-west1-a" areas of its Google Compute Engine service can now run apps directly on the Andromeda network virtualization stack. The search giant expects to migrate all the global sectors of the Compute Engine infrastructure-as-a-service to the architecture over time. (See Defining SDN & NFV.)

"Andromeda is a Software Defined Networking (SDN)-based substrate for our network virtualization efforts," explains Google distinguished engineer Amin Vahdat in the blog post. "It is the orchestration point for provisioning, configuring, and managing virtual networks and in-network packet processing."

Here's Vahdat's high-level diagram of the architecture:

Vahdat writes that Andromeda itself "is not a Cloud Platform networking product," but the basis for "delivering Cloud Platform networking services with high performance, availability, isolation, and security." The system was originally designed for Google's in-house use to provide isolated, high-performance networks with end-to-end QoS and availability. The company believes it has already deployed the largest software-defined WAN in the world with the Andromeda architecture. (See Google, Microsoft Challenge Service Providers.)

"Andromeda's goal is to expose the raw performance of the underlying network while simultaneously exposing network function virtualization (NFV)," he writes. "We expose the same in-network processing that enables our internal services to scale while remaining extensible and isolated to end users." Vahdat promises that a benefit of using the distributed architecture will be "major performance gains" as Andromeda spreads across the Compute Engine sectors in the coming months.

— Dan Jones, Mobile Editor, Light Reading

danielcawrey 4/5/2014 | 1:55:58 PM
Re: Packet Processor? This is very interesting - a case where Google has matured an internal technology enough to now sell to customers.

Both Amazon and Google seem to be at loggerheads in terms of bringing cool new stuff for IT departments to leverage. Things that can offer great preformance but at a lower cost than traditional architecture. 
dwx 4/4/2014 | 2:09:51 PM
Re: Packet Processor? What I gathered from the presentation from OFC Google did which you can find here:  https://www.youtube.com/watch?v=n4gOZrUwWmc

is it's more like a programmable ASIC tailored for higher touch packet functions at speed, not really a generic x86 processor or a dedicated appliance.  It may exist inside of an x86 server. You are talking about Google who have the scale and resources to develop their own dedicated hardware to make NFV more efficient.  

The underlay network from my understanding is still your basic IP/MPLS network running on common vendor routers.  
jabailo 4/4/2014 | 12:59:57 PM
Re: Packet Processor? Well for one, between two known parties inside a cloud, you might not need the whole tcp/ip envelop in a packet (since you know where it's going and where it's from).  So perhaps they can get efficiencies by reducing redundant data in each packet.

 
PoulAr 4/4/2014 | 7:19:29 AM
Packet Processor? Can anybody explain what kind of equipment a "Packet Processor" can be? Is it a x86 server or a network appliance or a service router? 

Google's claim for phenomenal performance seems to be enable by this function. It would very intersting to know a bit more about it and if anything can be applied to any NFV or CloudNFV activity.

 

Best regards,
P.

 
[email protected] 4/4/2014 | 5:27:40 AM
Re: picture Is that a serious message?

 
asdmognep 4/4/2014 | 3:01:14 AM
picture Dan,

with all due respects, the photo of you is horrendous. Please do humnaity a favor and switch it for something more flattering.
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE