IP Convergence: Beyond VoIP, Beyond Cost Savings

I attended the last day of the Carrier Ethernet Summit in San Diego last week, as part of my panel on Ethernet supporting Enterprise Services.

As I listened to what my peers had to say, such as Telcove and Yipes, I once again found myself puzzling over the statements they made in regards to Ethernet. I wanted very much to talk philosophically about ethernet as a carrier architecture, and whether all these extensions they want to add to it really make sense.

Unfortunately, my talk was already written. My three slides went like this:

Ethernet Myths

• Ethernet is established as a Simple and Easy Protocol in the LAN, so it makes sense to use it in the WAN
  
• Ethernet is more cost effective than SONET/SDH for the carrier
  
• VPLS is the latest and greatest way to do Ethernet
  
• Ethernet is the best way to support converged applications: Voice, video, and data.
  
• Ethernet means the carrier has a demark at the customer premise

Ethernet Truths

• Ethernet interfaces for routers are cheaper than their SONET/SDH equivalents
  
• Ethernet in the WAN is an afterthought, as evidenced by the many new standards currently in development (WAN-Phy)
  
• Ethernet does seem simple, but with all the extensions (802.1q, 802.1p, LAN-Phy, WAN-Phy) it is becoming even more complicated than the existing WAN standards.
  
• IP, not Ethernet, is the key to supporting the convergence of applications onto a single network

IP-VPN vs. Ethernet

• Access protocol and type (e.g. ethernet, SONET, DS3, E1, HDLC, PPP, SDH, Frame Relay, ATM) can be selected on a port-by-port basis
  
• non-IP protocols are typically already tunneled in IP for transport over the WAN
  
• Disaster recovery can occur with greater intelligence in an IP-VPN
  
• IP-VPNs are capable of providing superior Class of Service functionality
  
• Access to value-add applications such as VoIP, Video conferencing, VPN Remote Access, Internet Firewall and collaboration tools is much easier to provide in an all-IP environment

As you can see, a fair amount of this material was selected from a previous post I made.

One of the things that caught my eye during the Yipes presentation was a statement made by Keao about how customers demand the aboslute lowest in latency and jitter, and ethernet was best in both areas. He mentioned an application run by a financial services institution that required 250 microseconds of latency. Yes, MICRO seconds, not milliseconds... that's point-two-five milliseconds. Keao said that because ethernet didn't have the queuing requirements that a router had, that it was the only way to provide the level of service required.

I have three things to say in response to that:

1. That sounds like a very poorly designed application if it can't deal with at least 10ms of jitter. Even VoIP phones are designed to allow for more jitter than that (although more than 30ms will result in an increase in latency as a result of the increased size of the jitter buffer that is allocated in the end-points). Chances are good that this application will be redesigned over the next few years to perform adequately on an IP network, and the requirement for a super stringent data network will be moot.
   
2. The ability to prioritize ethernet frames a la 802.1p will increase the buffers in ethernet equipment re-creating the delay attributed (by a marketing presentation) to making IP slower
   
3. Looking at our SLA management tool (also available to customer now as well), it shows an average jitter of .31ms. I am looking at the path between Amsterdam and Tokyo and it is only .12ms, well within the tolerances of that crazy application.

Brian van Dussen from In-Stat, after referring to me as the fox in the henhouse, asked me what I made of the massive growth in sales of ethernet products (a bit tongue in cheek, I suspect). There is no denying that Ethernet is ubiquitous in the LAN and growing quickly in the MAN (probably due to the increasing affordability of dark fiber to the enterprise). The growth of IP-based applications as well as the convergence of legacy applications to IP is going to drive usage and expansion of the LAN and MAN infrastructures that support it, which is predominantly ethernet. One question that I have is how Ethernet is growing relative to IP growth...is it growing at a faster rate or equivalent? Can you even compare the growth in Ethernet equipment sales with sales of TDM and metro gear? Can you compare it with IP equipment sales? The complexity of the relationships between all of these types of technology suggests to me that such an observation, even if it can be stastically modeled, is meaningless.

I question is how smart it is to make the entire world one big Ethernet. It just seems to me that people are spending huge amounts of time and money to re-invent the functionality in Ethernet that the Internet Protocol already gives us, and yet ethernet will never replace IP, so what exactly are we trying to accomplish here?