A few weeks back, one of The Tolly Group's testing clients got a fair bit of press when it announced test results showing that its open source, Fast Ethernet router outperformed Cisco's 2821 Integrated Services Router.
"Twice the performance at half the price"
The emergence of enterprise-class, open source solutions is yet another chapter in a "commodity vs special-purpose" battle that has been raging for several decades now. Well on its way to becoming IT's version of The Thirty Years War, the struggle has gone hot and cold over the years -- but it has never stopped.
We've all experienced this struggle in one way or another. The vendor that has "something special" -- invariably described as "secret sauce" when it's briefing analysts -- wants to convince you that this special hardware is a key reason why you should buy its product instead of the competitor's, which uses conventional, off-the-shelf components. It is an interesting ploy, and one that often is successful when rhetoric rather than empirical information is used to win the argument.
Ask a reasonable person whether a specialised router crafted by a leading vendor with a no doubt substantial budget possibly could be outperformed by "free" software "thrown together" by a loose band of programmers. Should be no contest. Of course it was no contest -- but not the way you thought was going to be.
Granted, neither of these devices delivered wire-speed throughput at all frame sizes. And there certainly are devices we have tested that do. The open source entry did outperform Cisco's in every test, however. And "half the price" is really only half the truth. The open source offering is free. Its cost was related primarily to the Dell server that is the appliance the open source product uses.
When we ran this test, I couldn't help but think back to some of the earliest tests we ran -- back in 1991 -- involving WAN internetworking.. These were the first steps in corporate networks in the post-mainframe network era. Even then we saw such vendors as Cisco and Wellfleet (later Bay Networks and then Nortel) putting forth proprietary, special-purpose operating systems with proprietary I/O bus architectures. On the other side were vendors such as IBM and Microcom, using standard PCs as the basic bridge platform, to which they added standard LAN interfaces and a WAN card.
In the most stressful test of 13 products -- 64-byte packets across a T-1 -- IBM placed third with 2,486 packets per second (pps), with Cisco just behind it and delivering 1,836 pps. (And no, we are not missing "million" -- these were high speeds in 1991.) Cisco's entry was the MGS running a 25MHz 68020 and outfitted with proprietary LAN and WAN interfaces. IBM's was an off-the-shelf 16MHz PS/2 Model 55. This was based on IBM's ill-fated Micro Channel Architecture and had a standard LAN adapter with a specialised Vortex T-1/E-1 WAN interface. Although it looked like a PC running MS-DOS, in fact the IBM bridge program took over the hardware. MS-DOS was used only for keyboard and display I/O.
Marketing trumped technology, however. Competitors would say "You don't want to trust your network to a PC" -- even though it really wasn't a PC. And FUD prevailed. Today we can expect to hear the same comments made in reference to open source. Will they work this time?