17th May 2012

Routing Protocols and Computation in Silicon

I got this question and I guess it may not be obvious to everyone so I’ll have a shot at answering it.

Technology advances in ASIC hardware have resulted in substantial improvements in switching performances of routers and switches. However, the routing processes are still dependent on CPU speeds. What are the existing limitations in router/switch models which prevent route computations from being performed in hardware?

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

Problems With Cat6A Cables in Data Center

I was reading a white paper by Panduit that claims that 10GBaseT is suitable for use. I’ve been critical of Cat6A cable and believe that it’s not suitable for data centre use.

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

Cisco Nexus 5000 / 2000 Pricing Bundles and Fabric Extension Transceivers (FETs) vs 10GbaseSR SFPs.

Recently I noticed that Cisco is selling “Fabric Ethernet Transceivers” for the Nexus switch family. Some research shows that these are replacements for 10GBaseSX SFP modules. Importantly, it’s cheaper to install new cabling than to buy 10BaseSR SFP+ modules.

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

Fibre Connectors

A short summary of the Fibre Cable Connectors, description and some notes on usage. This is summary notes and intended for reference.

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

Soft Switching Fails at Scale

There is a significant camp of software developers who are developing software switching solutions for hypervisors. Which is nice, I guess. The use of software switching in the hypervisor has some good points but, in my view they are heavily outweighed by the bad. I present the use case, and show that software

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

Cisco C6500 Service Modules – Not My Choice, Now.

These all suggest that the time for planning and designing Service Modules is over. There are no suggestions that service modules for the Nexus 7000 will be developed that I can see. I can prognosticate that it would slow down the development of the core switch / route / performance functions, and it will be some years before those core capabilities is complete enough that service modules would become viable product development tasks — they might be in development, but not much chance of going into production. [^1]

Do I sound bitter about Service Modules ? A bit. I’ve had a number of hard to solve problems that lasted months before code fixes arrived. I’ve been fan of the NAM but the price is now far removed from it’s practical value. USD$30K List is way over priced for its capabilities and even with a 30% discount, you can buy a lot of network management systems that deliver much better functions and features for that price.

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

L2 MultiPath Basic Design Differences

With all the talk about Layer 2 Multipath (L2MP) designs going on, I just want to point out a fundamental change in the way many people approach network design. It’s seems that this point has been lost somewhere in the discussion of protocols.

The Spanning Tree Protocol blocks looped paths, and in a typical networks this means that bandwidth is unevenly distributed. Of course, we might use PVST or MST to provide a rough sharing of load by splitting the spanning tree preferences for different VLANs, but the design still doesn’t change overall. The basic point is that there is a LOT of bandwidth that is never evenly utilised – and that means wasted power, space and cooling (which costs more than the equipment itself).

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

VMware: Let’s Get Logical – The Case for OpenFlow Network Virtualization (and Their Failed Network Plans)

VMware has made several strategic moves to implement dynamic networking – vSwitch, vDS, Nexus 1000 (in partnership with Cisco), vCloud External Networks (using MAC in MAC of all things) and have basically failed to deliver overlay technology without implementing technology in the network itself. Equally, VMware hasn’t been willing to engage with the networking vendors to develop technologies that would solve this problem – VNtag / VEPA/ VEP combined with TRILL / SPBB, instead letting them argue amongst themselves. VMware attempt with vCloud networking using MACinMAC encapsulation seems to have failed and stalled and is getting another attempt using MACinIP. VMware/Xen/HyperV are all desperate to have a more dynamic network that can be controlled from their software and this might be where OpenFlow gets a big lift – as a configuration engine.

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

PacketShaper and Flow Directions

I stumbled across an old diagram I made a long time ago about the direction of flows on a BlueCoat PacketShaper. Since I’ve been looking for it for about three years, I’ve diagrammed it quickly so that it is here for future reference when I’m working PacketWise in the future. PacketShaper PacketWise is one of my very favourite tools for managing traffic flows, and much preferable to PHB QoS aka DiffServ for many types of use cases.

An TCP flow has four possible directional attribute related to the use of a inside and outside networks, and whether the flow was initiated from the client to server which sets the “direction” of the flow relative to the Packeteer. The flow is determined by who initiated the three way handshake. For purposes here, the Client always initiates the TCP connection, and the Server terminates the connection.

TCP Session and Direction

Most people understand the three way handshake, but not many consider the direction of the session.
Packet shaper flow directions 0

The connection from the client to the server is outbound, but is inbound on the server. And vice versa, the server outbound session is inbound on the client.

Packet shaper flow directions 0 1
That’s not very useful for being able to define the direction of flows.

Why is direction important ?

Direction of flows is important if you want to configure asymmetric rules. That is, not all protocols require symmetic bandwidth. For example, HTTP traffic is usually a 10:1 ratio for reply to request. That is, a request for this webpage is about 10KB, but the reply with the data, images and javascript is more than 100KB.

Packet shaper flow directions 0 2

For an FTP upload server, you might have the reverse condition where the inbound traffic is far more than the outbound.

To make the most of your Internet connection for this case, you could configure the inbound bandwidth on your Internet connection to be 80% FTP, 20% HTTP and the outbound bandwidth to be 20% FTP and 80% HTTP. This gives a far better utilisation, especially in regards to better TCP Windowing and overall TCP goodput.

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved

Complex Systems Have Complex Failures. That’s Cloud Computing.

Exposing cloud failures The result of the Amazon EC2 failure this week has exposed a number of technology strategies in cloud infrastructure as being less than perfect. Complex systems have complex failures The most vexing problem of Cloud Computing is that these systems are complex, and the more complex system the more complex the failure. [...]

This post is copyright of Thropos Ltd ©2008-2011 at Etherealmind.com - contact | email: greg.ferro@packetpushers.net - twitter: @etherealmind | All rights reserved