Wednesday, September 12, 2007

Is Ethernet The Right Solution For Your Data Network?

This is not a trivial question. Ethernet is relatively ubiquitous, in that most computers and peripherals speak ethernet. Ethernet is routable - far too many protocols aren't - and is relatively fast (10 gigabits per second for a high-end card, but you can bundle up to 10 of them to create a pseudo-100 gigabit connection) and is relatively cheap.

Ok, those are the benefits. Are there drawbacks? Yes. Ethernet isn't good on latency, there's no meaningful error correction, there are many defined types of ethernet frame (which means you can assume companies have taken shortcuts in testing), offloading of security and checksumming is rare to non-existent, very high-speed cables are amazingly fragile, hardware multicast support is generally minimal (a bother for LANs where many protocols are multicast these days) and vendors often cut corners on using components of adequate quality.

Ethernet definitely is the way to go for internal data systems. Now connecting two buildings together, not so much. Ethernet has a length limit so it's best for use in connecting all of your machines together to a switch or switches and from there you can connect to the internet or another building using some other form (T1, DS3, OC3, really depends on how big the company is and how much bandwidth is really needed).

In addition, most Telecommunication and IT professionals have had to work with hundreds of Ethernet connections (monitoring, troubleshooting, testing, etc.) but almost none of these people have had to troubleshoot a T-3, POS or ATM link. This shifts the knowledge advantage to the carrier in case of any issues, as they often have more experience with those technologies than any end-user. You often have to take their word for it that the issue is your equipment. When you can swap your equipment in a second (potentially with your laptop) you get on to the next troubleshooting step much faster.

Are there any alternatives that are good? Well, Infiniband is routable and can be used on both LANs and WANs now. It has (almost) none of the problems of ethernet and is therefore in an excellent position. The downside is that it has (almost) none of the strengths, either. It's rare, vastly more expensive and the odds of finding any COTS peripherals that support it is essentially zero.

The latest round of wireless networks are almost as fast as ethernet, eliminate the cabling problem completely, but don't play nice on crowded networks and have nightmare routing issues if multiple access points are involved and you're wanting to have true mobility. (Mobile IP and Network Mobility are experimental, rarely implemented and even more rarely implemented well.)

In the end, the "best" business solution will involve some hybrid solution that combines different networking technologies, such that you mitigate as many of the drawbacks as you can (by not using a solution where it doesn't fit) and harvest the most of the strengths as you can (by using solutions specifically where they fit the best). You will never find a single one-stop solution that fits all cases, you will only ever find one-stop solutions that fit some specific part of the problem.

One-solution shops are invariably the places that have the greatest problems, the least reliability and the most headaches, although they are also the places that have the lowest maintenance overheads because they don't need to have the additional expertise to hand. However, IT costs should not be seen in isolation, but in terms of the net cost to the business as a whole. A "cheap" answer that harms the company is more expensive than a "high cost" tuned answer that benefits everyone.

From a business perspective, how do you balance all of this and make a sensible choice? If I knew the answer to that, I'd be rich and living a life of ease. The best I can tell you is that there are techniques for solving such problems for specific cases. Probably the best-known method is SIMPLEX, though there are many others that are probably better-suited. Define the resources and impacts, then solve the inequalities to give you a best business case. Hiring a consultant isn't going to give you anything much better - maybe some but not much - but will cost you considerably more in time and money.

The bottom line (pun intended) key reason for the deployment of Ethernet as a WAN technology is cost. We are only talking about the last mile here (and sometimes the last few meters). The actual transport network is generally POS or ATM from the providers standpoint. I have seen 10GB links that are actually provided over an OC-192. But from the CPE standpoint, the costs are simply buying the most ubiquitous interface for that speed.

For assistance in finding the best fit ethernet solution for your business application ... for both cost and function .... I suggest the free services offered here [they also cover DS3 and OC3 bandwidth networks] :

DS3 Bandwidth, OC3 Bandwidth, and Ethernet Solution


Labels: , , , , ,


Post a Comment

Links to this post:

Create a Link

<< Home