≡ Menu

When MIPS are free

What would you do with thousands of “free” MIPS?  In The Rise of “Freeconomics”, Chris Anderson notes that for about $200, you can buy an Intel Core Duo processor, capable of executing 20,000 MIPS.  To put that in perspective:

  • In 1977, Digital Equipment’s Vax 11/780 was a 1 MIPS minicomputer, and the Cray-1 supercomputer delivered blindingly fast execution at 150 MIPS.
  • By 1982, 5 years later, a 6 Mhz 286 had about the same equivalent processing power as the Vax. 
  • Sometime in the mid 1990’s, Cray’s benchmark was finally passed on PowerPC processors, as PowerMac’s emerged benchmarked at 150 to 300 MIPS.
  • A 1999 era Pentium III/500 delivered 800 MIPS of processing power.
  • A year later, in 2000, the Playstation 2 pumped out an astounding 6000 MIPS.
  • My 2002 vintage Athlon XP clocks in at 4200 MIPS.
  • And today, for about $200, you can buy a 20,000 MIPS processor. 

Current embedded processors (like the PXA900 in my Blackberry Pearl, or the ARM 1136 in the Nokia N93 I wrote about) are capable of 2000 era desktop processor speeds — in the range of 1000 MIPS, depending on battery consumption. These devices take photographs, play music, perform reasonable quality speech recognition and much more. 

Our office runs a year 2000 vintage 700Mhz Pentium III with 256M of memory and Asterisk for our PBX. We can accomodate perhaps 6 or 8 phone calls simultaneously (there’s just 12 of us, so that’s really all we need).  In theory, my current cellphone, and my current PC-based PBX, have roughly the same processing power.  It should be possible to power an office PBX on a PDA. 

It’s 2006 now.  If the current trend holds true, and we can each carry 20,000 MIPS of processing power in the palms of our hands by 2012, what will we do with that power? Will my phone be a peer to peer media player, with advanced 3D graphics, and high definition video? The processor in my router is a 266Mhz part from Broadcom.  What will the world look like when thousands of MIPS of processing power are available at the edge of my home? Does my router become my home web page, PBX, mail server, and… who knows? PC’s will be valuable for their keyboards, but most of the interaction you have with computers will be via portable and embedded devices.

One of the biggest impacts will be on network architecture.  Isenberg’s The Rise of the Stupid Network posits that all intelligence will move to the edge.  In a world where MIPS are free, it’s simply more cost effective to distribute the decision making to the edge of the network rather than centralize it.  The implication for network service providers (especially telecommunications companies) is that the centralized applications they relied on for revenue in the past are going away.  You can already see this happening.  Why, for instance, would I ever type *69 into my handset to ask the network who the last person who called me was when my handset keeps track of the last 50 callers, and lets me punch a button to redial any of them.  When *69 was conceived of, telephone terminals were dumb.  Today, they’re not.

Only resources which must be centralized will remain so. Directories will likely remain central assets, because decentralized directories are hard to manage, as Skype has shown repeatedly.  How often have you picked up a real phone, and called a Skyper to ask if they were online, because they didn’t show up in your buddy list?  The value of a social networking application is in the relationships it represents.  Centralized servers will be required to manage these applications.  There will be tremendous value in a few key centralized services, and everything else will be decentralized. 

One thing is for certain.  When everything is connected, and processing power is “free”, today’s communications networks will seem as antiquated as steam locomotives must have been after the introduction of the automobile.  That day isn’t that far away. We will see it within the five years.

{ 7 comments… add one }

  • Christopher Small November 27, 2006, 11:43 am

    "In 1977, Digital Equipment’s Vax 11/780 was a 1 MIPS minicomputer […]" In fact, the VAX 11/780 executed about 500,000 instructions in a second. This was not widely known at the time, even inside of DEC. http://portal.acm.org/ft_gateway.cfm?id=285946&am

  • Salim AbiEzzi November 27, 2006, 5:34 pm

    I disagree with the conclusion that free processing power will automatically mean a reduction in centralized services. Another dynamic that is starting to occur is the advent of gigabit pipes to the premise, which will render centralized services/storage as accessible as local disks. This dynamic opens the door to offloading all headache associated with maintaining a PC (software installation and upgrades, virus protection, spywear, backups, etc) to centralized services that remote the whole UI to HDTVs and productivity high-res monitors at the premise; not even the browser is running locally. This approach has the added advantage of location transparency of services and user data including digital memories (photos, video, etc).

  • Alec November 27, 2006, 7:21 pm

    Hey Salim — nice to hear from you! I think what you've described is a usability flaw in the current generation of technology, rather than a fundamental argument against decentralized processing. In a world where storage is effectively infinite (heck, I've got a terabyte of disk around the place right now…), and processors are cheap, connectivity, even with gigE to my doorstep, is still expensive. As the vendor supplying all those "free" services, I will want to push as much of the expense to the user as I possibly can.

  • Moshe Maeir November 27, 2006, 9:51 pm

    Alec, I tend to agree with Salim. With the advent of web applications such as the suite of Google applications the user doesn't need a smart terminal! Reminds me of the days when I started out in computing in the '70s and every time we wanted to do something – it was off to the computing center with a pile of punch cards. The IBM mainframe did all the processing. It is almost the same today, hosted web apps replace the mainframe. Due to the ubiquity of broadband – I can access the server from anywhere, no neesd to shlepp over to the computing center :-)

    Personally I don't carry around a laptop anymore, just my Nokia E61 which gives me great email and web access, sure it is "smart" compared to the punch card terminal, but it is still a "dumb terminal" working off the web servers. If connectivity is "free" why would I want to carry my data and applications around with me?

  • Alec November 28, 2006, 3:00 am

    Moshe, I think you guys are both still thinking desktop. Your E61 is a remarkably powerful device. It's the equivalent (in terms of processing power) of a 2001 era PC. What you're doing with that is 2001 era client-server computing (at least for email). Would you seriously replace all the capabilities of that device with an over the air substitute? It may be true that high speed fibre to the home is becoming real, but mobile bandwidth isn't at GigE speeds and may never get there.

  • Saad Shakhshir November 29, 2006, 1:11 pm

    I have to disagree with you Alec. Even in the mobile space, I do not see the long term trend going towards having the intelligence lie on the device. The trend is clearly towards having thin clients and centralized databases and application servers. Why should I pay for having a complicated piece of software that I have to maintain on my device when I can have the same functionality through a web connection to a server somewhere? That server holds all my data and guarantees me a consistent experience over a variety of devices. It also means I don't have to replicate the software on the various devices that I own, like my mobile phone, my laptop, my desktop, etc. Saying that mobile bandwidth will never get to Gigabit speeds when you have just clearly pointed out how Moore's Law isn't slowing down, is just inconsistent. Ten years ago we were all on 28.8k modems, and that's if we were lucky. Now 10Mb/s is easily accessible, a 500-fold increase. Never say never, especially when it comes to technology.

  • Alec November 29, 2006, 6:35 pm

    Saad, the bandwidth problem is a different problem from processor speed. Spectrum isn't like silicon. In silicon you can make the die larger, or pack two or more cores side by side. Spectrum is fixed. Jumping into higher frequency spectrum (say 5Ghz vs 2.4Ghz) increases the amount of data you can carry, but at the expense of higher power requirements and decreased range.

    Bandwidth is likely to continue to lag storage and processing speed, in my opinion, for the foreseeable future.

Leave a Comment