& cplSiteName &
Comments
Newest First  |  Oldest First  |  Threaded View        ADD A COMMENT
dwx
50%
50%
dwx,
User Rank: Light Sabre
2/19/2016 | 5:17:16 PM
Re: Is this good for the planet?
The datacenter is no different from any other peak-capacity designed element.  It will be busy during its peak times and idle off peak.  There is functionality built into routers these days to shut off unused ports, ASICs, etc. if you want to during idle periods.   

Broadcom sure, that's an NPU, but once you start throwing NPU functionality into a server, is it really a server anymore?   In the end we'll likely end up with NFV-optimized devices which are pretty much the same as the virtualized firewalls, routers, etc. we've had for years now. The trick is it being commodity hardware like we've seen with white box switches to keep the HW component cost low and the HW being open to run different operating systems on the HW.     

The reality is the power/space requirement today for packet processing on generic x86 is still many times what it is for dedicated hardware.   It will undeniably get better over time, but it hasn't significantly improved in the last several years.  Of course there are some applications where it makes a lot of sense, where you want flexibility in the purpose of the box for lower speed customer functions, CPU-intensive control-plane tasks, etc.    

 

 

 
dlitvine
50%
50%
dlitvine,
User Rank: Light Beer
2/19/2016 | 4:50:11 PM
Re: Is this good for the planet?
Now start thinking of data center which benefits from stat multiplexing and almost 100% utilization, optimized redundancy scheme etc. Compare it with very common use case of fully redundant router which is forwarding at 10% of its capacity but chewing power like a thirsty horse.


This is where industry going to. Of course there are corner cases where you still need special silicon to get throughput and efficiency. No doubt, but merchant silicon vendors investing heavily in NFV,  this is matter of time when intel/broadcom chip for server and LAN switch will be on par.

 
dwx
50%
50%
dwx,
User Rank: Light Sabre
2/19/2016 | 4:06:57 PM
Re: Is this good for the planet?
.75 is for a whole system, the actual line cards are starting to approach .50.  Of course this isn't for a system with the fans running full speed because they only run full speed during some kind of catastrophic cooling failure.  Of course higher density systems benefit because they only have one set of commons.  

I have a MX80 which is built on a chipset that's 4+ years old at this point but is a 80Gbps full-duplex system.  At full tilt it consumes around 200W of power.  

I have a new Dell R630 server with two 8-core processors, much less powerful than the one they used in their dataplane server.   Running 20G of traffic through the server it uses about 260W of power.  

So in that real world example I have an old MX80 using about 2.5W/Gbps and that includes terminating services, firewall filters, ACLs, etc.    If I tried that on the R630 it wouldn't be able to sustain 20Gbps, and even doing nothing but forwarding traffic it's using 13W/Gbps.  

 
dlitvine
50%
50%
dlitvine,
User Rank: Light Beer
2/19/2016 | 2:40:45 PM
Re: Is this good for the planet?
Your calculation is incorrect.

1. When vendor is claiming 0.75 watt per Gbps, this is chip efficiency for line cards. It doesn't take into account RP/Fabric/Fan consumption and of course these marketing numbers are using "best" exampe for very high density.
2. Power supply nominal value had nothing to do with real server power consumption.
It can be times lower.

If you look at the minimum Cisco ASR9904 configuration (just one line card and one RP) you will see the consumption of 500-600 watt and 3kwatt power supply required.

Doesnt look much different from your example, right? No matter what marketing numbers are.
dwx
33%
67%
dwx,
User Rank: Light Sabre
2/16/2016 | 11:40:10 AM
Re: Is this good for the planet?
The datapath server using the $4000 CPU is probably using at least a 800W power supply but it's hard to say how much it's drawing.  So if you are looking at W/Gbps, if you use 80Gbps you are looking at maybe 10W/Gbps.  A modern service router like the Cisco ASR9K, ALU 7950, or MX is around .75W/Gbps in its latest iteration.     

One recent slide deck I saw last year from Cisco put the space/power utilized by a x86 setup to be about 15x more than equivilant dedicated hardware and that's probably still accurate. 
mrblobby
50%
50%
mrblobby,
User Rank: Light Beer
2/12/2016 | 10:45:34 AM
Is this good for the planet?
How do VNFs and their platforms compare to proprietary machinery in terms of watts per Gbps?
Ray@LR
100%
0%
[email protected],
User Rank: Blogger
2/10/2016 | 1:00:38 PM
Deep dive on the webinar
There is an upcoming webinar that's going to get deeper into the tests process and outcomes - anyone can join and ask questions of the presenters -- check out the details at

Evaluating the performance of Nokia Virtualized Service Router (VSR) and Virtualized Mobile Gateway (VMG)

http://www.lightreading.com/webinar.asp?webinar_id=617

Wednesday, February 17, 2016, 11:00 a.m. New York / 4:00 p.m. London
Helen80
100%
0%
Helen80,
User Rank: Light Beer
2/10/2016 | 3:46:07 AM
There you go Nokia
It seems the company has some superior virtual network capabilities.


Featured Video
Upcoming Live Events
September 17-19, 2019, Dallas, Texas
October 1-2, 2019, New Orleans, Louisiana
October 10, 2019, New York, New York
October 22, 2019, Los Angeles, CA
November 5, 2019, London, England
November 7, 2019, London, UK
November 14, 2019, Maritim Hotel, Berlin
December 3-5, 2019, Vienna, Austria
December 3, 2019, New York, New York
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events