Category Archives: Network Economics

Network Switching Costs II: The 55% Tax (B)racket

Posted on by
2

In a previous post, we looked at trends of cost per port for data center network switching.  Now, everyone is focused on cost per switch port but nobody seems to be paying much attention to the fact that every port requires a transceiver and a cable (whether optics or direct-attach copper).

Let’s look at your run-of-the-mill data center deployment to wire up three thousand odd servers at 10 gig. Pick your poison, but basically the transceiver-cabling tax % overhead stays the same, independent of the vendor.

Network Switching Costs

Server Access Ports (10GE) 3072
Access Uplink Ports 1024
Total Access Ports 4096
Interconnect Ports (to Aggregation & Core switches) 2560
Access ASP/Port $280
Interconnect ASP/Port $700
Access Switches ASP $1,146,880
Interconnect Switches ASP $1,792,000
Total ASP Switches (1) $2,938,880
Server Access DAC Cables 3072
Interconnect Transceivers 3584
DAC ASP (3 meter) $60
Fiber Transceiver ASP (SR) $400
Server Access DAC Cables ASP $184,320
Interconnect Transceivers ASP $1,433,600
Total Cables+Transceivers (2) $1,617,920
Cable + Transceivers Overhead as % of ASP Switches Cost(3=2/1*100) 55%
TOTAL $4,556,800

Welcome to the 55% vendor tax.

And, where else might you have expected to pay an average effective tax rate of 55%?

In the Soviet Union. In 1990.

(page 33: http://www.imf.org/external/pubs/ft/fandd/1996/06/pdf/cheasty.pdf)

Glasnost? Niet. Support for apparatchik transceivers only.

Surely, you say, a lot of R&D investment has gone into these gizmos to justify their price? Nope, all made by the Avagos, Finisars, Opnexts et al. of the world.

And your vendor can publish thousands of RFC’s and other protocol alphabet soup in their datasheets but not a list of compatible (=the ones that they will support) transceivers??

So if you would like to see at least one of your favorite vendors sing and dance the kalinka, be sure to request the cheaper USR transceivers* for the 100m runs so common in today’s data centers :-)

*: A discussion on USR transceivers at Greg Ferro’s blog.

Network Switching Costs I: The Economics of Open Networking

Posted on by
7

There have been a lot of really good discussions and conferences this year (see Woodstock moment at www.opennetsummit.org) on the promise of open, software-defined networking (SDN) to end the Tyranny of the Network — too expensive, too closed, too much work, too static, too insecure, too many VLANs, too few VLANs, too many ARPs, too few MACs, too many pages in RFCs, too little content, too many protocols, too slow innovation, too much FUD.

As such OpenFlow and SDN have momentum on their side.

And there is no question it leads to some really interesting academic work like formally provable characteristics of the network (view Nick McKeown’s talk video here), new programming languages (www.frenetic-lang.org), students being able to experiment with new protocols in the wild (www.geni.net).

But what’s the deal commercially?

In other words, what is the economic value created by merchant silicon and open networking software?

It turns out people will buy more at lower prices and vendors want to sell more at higher prices.  So price and quantity demanded are inversely related and that’s called the Demand curve. The Marginal Revenue is the incremental revenue that is up for grabs for producing one additional unit of the product. Without getting too Econ101, the slope of the Marginal Revenue is twice as steep as that of the Demand curve. The only other curve we will contemplate here is that of the cost to produce one additional unit of the product (Marginal Cost).

OK, what’s the point?

The point is that the quantity that sellers produce is dramatically different whether a particular market is a quasi-monopoly or perfectly competitive. And if we were to talk about the data center and enterprise Ethernet switching market, today its state is much more the former than the latter.

Exhibit 1: State of the switching market today: a textbook case of a quasi-monopoly

Exhibit 1: State of the switching market today: a textbook case of a quasi-monopoly

Exhibit 1 shows the state of the switching market today, a textbook case of a quasi-monopoly.

In a quasi-monopoly, the quantity produced by the sellers is Q1 (where MR intersects MC). The price of that product is P1. In the case of, say, data center top-of-rack switches, P1 is about $14,000 today (give or take for your 48 to 64 10gbps-port job). Assuming cost of goods sold (COGS) is about $3,500 and that marginal cost is double COGS (leaving why as exercise to the reader for a moment) MC = $7,000.

Exhibit 2: State of the market as it would be with open networking: ~=perfect competition

Exhibit 2: State of the market as it would be with open networking: ~=perfect competition

Exhibit 2 shows the state of the market as it would be with open networking — perfect competition.

Perfect competition means free entry and exit, and as industry barriers go we will get pretty close courtesy of merchant  silicon in switching, contract manufacturing, open source everything, i.e.,  (formerly) fixed cost, say hello to variable cost. Now the sellers produce at quantity Q2.  And here is the kicker: Q2 = 2 x Q1, because it lies at the intersection of D and MC. Assume MC about flat in the range. (Note: MC and COGS remain about constant over time but obviously each product vintage will bring port count/speed bumps following Moore’s-or-other Law). Lo and behold, we are buying the same (substitute) top-of-rack switches for P2 = $7,000.

This isn’t to say every switch will go for an average selling price (ASP) of $7,000. In fact, a much richer ecosystem will see more switch makes, more flavors of how, where and with what performance switching is done in the data center, more choices of bells and whistles, and maybe more business models in how the stuff is consumed. But the market’s willingness to pay for today’s “state-of-the-art” switching function proper (meaning data center-class L2 + basic L3 feature set) will be $7,000 — roughly half.

If we call the Ethernet switching market about a $16 billion run rate then the consumer surplus that open networking will create is worth $4 billion per year (the area of the yellow triangle). Using an enterprise value/revenue multiplier of 3 (as of today Dec. 1 2011, CSCO is a bit under 2, JNPR a bit over 2, RVBD a bit over 5, and FFIV a bit over 7), this values merchant silicon + open networking value creation at $12 billion.

The question then is, with this newfound $12 billion consumer surplus, what innovation will he or she want to spend it on?

So Open Networking promises more diversity, faster innovation and a richer ecosystem — but if all else fails we still get to buy two switches for the price of one. And if public cloud is to be 50% of their market by 2015, this is one stimulus package we all share in.