Enterprise Technology By Phil Edholm

Older: VideoConferencing - Going Consumer?
Newer: Ferrari versus Corvette - Which would you Choose?

Is There an Internet Traffic Jam Coming?

By Phil March 6, 2008 1:39 pm

I started to reference the issue about the question whether the Internet bandwidth was going to become an issue in a post response last year. At the time I referenced a report issued by the Nemertes group that predicted that the capacity of the Internet would become an issue between 2010 and 2012. The post I did about Microhoo and Google competing to introduce new services and capabilities began to highlight this issue in my mind as well. Then last week I had a meeting with a key technology executive from a major North American service provider who indicated their bandwidth is growing 40% per year and they have begun the migration to 40 Gbps in their fiber backbone.

The combination of these three events led me to revisit this critical question; "will the inability of the network to deliver bandwidth become an issue?" If you read the Nemertes report entitled "The Internet Singularity, Delayed: Why Limits in Internet Capacity Will Stifle Innovation on the Web", it details a view that both the core of the network and the edge (especially the shared bandwidth at the edge) will not grow as fast as a projected demand growth. Essentially this model predicts that somewhere between 2010 and 2012 the available bandwidth will not keep up with bandwidth demand growth. If you refer back to the discussion of bandwidth in I did last year, this will coincide with the critical video transition in the network. If we map 40% growth from today forward and assume that the carriers are moving to 40 Gig backbones over the last 6 months due to capacity demands, hen sometime in 2011 the 40 Gig fiber solutions being installed today will be at the same load points as the current 10 Gig solutions were last year. This would mean that to meet the 2012 demand, 100 Gig fiber solutions are required (or additional fibers). Of course, this all caries out into the metro and access networks.

bw-growth.gif

On the surface, it would appear that the natural growth of bandwidth, combined with the transition to video is raising a distinct possibility that sometime between 2010 and 2012 there will be an inflection that, unless driven by new technologies and capabilities, will begin to impact overall service levels. The critical question is whether this event will coincide with new apps and services that can only be realized if unencumbered by such a "slow-down". Would the take-up of YouTube be as dramatic if the performance was highly variable and often degraded significantly. Would the new user community wait if the variability for downloading a video was very high, occasionally taking minutes instead of seconds. It seems to me that this concept, combined with the explosive generation of services that a Microhoo versus Google competition may engender, has a high probability of creating a period of instability and reduced expectations. This is both a challenge and an opportunity, but will only be realized as an opportunity if we begin moving forward with technologies that increase bandwidth and simplify the network switching environments.

Categories: Networking, Availability, Unified Communications, Technology, Trends, Architecture, Uncategorized Trackbacks (0)

Tags: No Tags

Comments

  1. Phil, Thanks for citing our report. Since we generated the original model, we have heard lots of anecdotal information from various sources that reenforce our original conclusions. There are lots of creative people out there inventing things that will consume bandwidth on the Internet. Thanks to people like you, we are begining to think in terms of how we will address those bandwidth demands.

    March 6th, 2008 at 7:25 pm by Mike Jude

  2. Phil,

    Interesting article if only for the point that you yourself have openly argued that bandwidth will always out strip demand, meaning QoS is an overcomlicated and unnessasry technology.

    If the future is no longer 10gigabit but 802.3ab and it’s 40gigabit and 100gigabit standard, where does this put Nortel in meeting the demands of the futue when it struggles to today with 10-gigbit density!!

    Nortel Observer

    March 8th, 2008 at 4:56 pm by Nortel observer

  3. While I am not sure if I have done it here on the blog, I have indicated that network bandwidth did not keep up with Moore’s Law for the 80s and most of the 90s. What happened is that 2 of the three BW types (core optical and LAn) grew faster that Moore’s Law from about 1996 to 2006 (LANs went from shared 10 to switched gig and cores from 2.5G single lambda to 10-40gig DWDM with up to 160 lambdas). What did not keep up was access bandwidth (edge to businesses T1s and E1s versus 10-100M). That is changing now. However the biggest change is the 10-20x (and 30-100x with HD) increase in BW that video is having. As I indicated in the post referenced above, this is a major change in BW. However, the point about QoS still applies. As most of the video is cached at some level it is not real-time, therefore the capacity create enables true real-time (voice, videoconf) to use simple CoS and statistical capacity. Simple priority and never discard queues for the real-time traffic should be sufficient.

    However, the real issue is will the network run out of total capacity for all the “best-effort” use apps such as YouTube (and the next video generator) or other next generation applications that use increased bandwidth.

    As to the migration to 10 gigs, it is not an enterprise campus question at all. For a campus with 5,000 users and 50% active at any point in time, we would have 2,500 users. This is a large campus by size (probably only 1% of campus networks are this size or larger). If we assume 80 Mbps per user (this equates to stereo HD with full side screens at each seat creating an immersive experience), we would need 200,000 Megabits or 200 Gigabits. If we assume today’s Nortel 8600s in a dual S-MLT configuration, that is 100 Mbps per switch. With a backplane capacity of 720 Mbps (OK, that is Cisco math, it is really 360), this is 28% of capacity. Obviously this is worst case. This type of desktop set-up would cost around $10-20K for 3 monitors, the 3D rendering software and the necessary 3D LCD glasses for the stereo display. A more reasonable analysis would be 100% of desktops running progressive scan 1080P video with a bandwidth of about 24 mbps. Thi yields a BW per switch of about 60 Mbps or 16% utilization. Even in a switch failure the demand on the resilient switch is only 32% of capacity. Of course both of these scenarios dramatically overestimate BW. If we assume 40% actual PC use time on average and 30% for output at 1080P (the otehr 10% for input at less than 300 bps typing or for 1080P video the other way…), then teh utilization drops to under 10%. If we double the users we can put 4 switches in parallel for 2x more capacity. With 2-4 10 gig uplinks per wiring closet and 20 wiring closets, this is only 40 10 gig ports per switch. So the issue is not the campus. The issue is in the data center and the carrier core. In the data center you need more 10 (and maybe 40) gig connections primarily for blade servers. If you are going with 1U servers, the top of rack with 1 Gig verticals (note there are actually 2 per server for 2 Gigs)works as it matches the internal bus speeds of the servers. For blades that have 14 server cards, you need 10 gig for the load that the 14 dual processor plugs with quad cores could provide (I think this is 112 processors per blade server).

    The core carrier network is where capacity will be challenged as discussed above. with all this new apps and video bandwidth, combined with the rapid scaling of access bandwidth through Metro Ethernet in the enterprise and next gen cable and carriers (both copper and fiber) to the residence, teh demands on the core could outstrip both Moore’s Law and the ability of the optical core networks adn the routers to keep up.

    As to what Nortel is doing…we are scaling our switching for the data center along with driving the best top of rack solution in the industry (the only one endorsed by VMWare), we are continuing to improve our 10 gig density in the campus core, and we have introduced 40 gig optical that enables carriers to use fibers that are in the ground but not capable of going at higher speeds without modulation. And we are moving fast in 100 gig and 4G technologies.

    The comment about wireless is interesting. I have been noodling whether wireless impact will change the dynamics in a new way. Obviously, wireless movement to 4G is required to have the impact on the core that would be anticipated by the 3 Billion wireless users. The following is a quick calculation comparing wirelss bandwidth to wireline with the 10:1 relationship predicted in the Law of bandwidth.
    1. There are about 3-4x more wireless nodes than wired nodes (3+B to about 900M)
    2. If wireless speeds are 10% of wireline (Law of bandwidth)
    3. If 50% of the wireless nodes are “smart”
    4. The percentages for wired and wireless that are active with the same apps is similar

    Then there are 2 times as many wireless nodes as wireline, but they are using 10% of the bandwidth each, so the wireless user base contributes a 20% increase in BW over the wireline base.

    However, if 3G is compared to next generation wireline Cable and VDSL, the difference is more than 10x. If 3G is 300-500 Kbps and wireline is 30Mb to 100 Mb, then the difference is 100x (this is because the 3G step is one generation back in the model). In that case the wireless contribution at the core is less than 5%, more like 2%.
    So in the end, what is changing is the advent of video (n addition to Moore and Metcalf) and the capacity at the access networks in both residential and commercial offers to enable the utilization of the 5-10x bandwidth increase.

    Sorry for the long response….lots of thoughts here…..

    March 11th, 2008 at 8:29 pm by Phil

Leave a Reply