Friday, 27 April 2012

NBN Internet traffic drivers: STORAGE


Since IBM introduced the first disk (5MB, RAMAC) in 1956, demand for storage has grown relentlessly, as demonstrated by yearly total disk sales. Disk Vendors have collapsed to 2+1 as we enter the end-game.

Storage drives demand for bandwidth, both for data exchange/transfer and backups. Bandwidth demand will track storage sold, lagged by 1-3 years.
The proliferation of smartphones with video has driven the explosion of content on social media, and You-tube's "60 hours of video uploaded every minute" is only set to get higher.

By 2020, the last 10-fold scale-up of disk drives will be complete with 3.5" drives maxxing at 20-40TB: 1-10,000 hrs video [or for 'action' HD, ~500 hours with MPEG-4 variable encoding/bitrates.]
Flash is also in its last 10-fold scale-up and starting to suffer the same scaling problems that began in 2004/5 for DRAM. There are at least 3 candidate technologies to replace Flash and fuel further growth.

The bandwidth driver for domestic Internet will be driven by upload speed (currently 1-40Mbps), not the quote headline rate of '100Mbps' for download.

This has 3 components:
  • file backups,
  • content sharing and
  • real-time low-latency access (telephony, video calling and gaming, ...).
At the mid-range 5Mbps uplink, video uploads will only be 2-3 times faster than their record rate, but only if the link is not otherwise engaged in things like Skype and gaming.

As families realise they've moved to entirely digital records of their lives and that disks fail and/or lose data unpredictably, they will start to subscribe to Digital Preservation services:
every day they will automatically backup masses of documents, images and video to "the Cloud".
Because of the shared uplinks of the 'PON' [Passive Optical Network], even this promised line-access-rate will rarely be achieved. PON Congestion during busy hours will be considerable - exactly the same as the Cable TV (HFC) networks suffer now. This is on top of the ISP's 30-100 times over-subscription rates.

By 2020, there will be huge demand for Gigabit PON to be enabled. Not for the download speed, but for the 50-400Mbit/sec upload speed. As more people opt for the higher upload-rate plans, the "last mile" of the network, the most expensive part to change, will become more stressed.

Simplistic analyses based on "raw access speed" usually contain another two major flaws:
  • latency (round-trip delay) and jitter are far more critical for real-time services than raw throughput.
  • utility networks need to cater for peak demand, not average demand.
The lesson from the electricity networks is clear.
While homes on average consume under 1kW (20kW-hrs/day) during busy hour" demand rises to 3-5 times that, with instantaneous household consumption 10-20 times the average. Consumers will pay a premium for just 10-hours/week of higher-speed uploads.

For home "power users", Petabyte storage (1 Million Gigabytes) will be within reach well before 2020 - technically and financially. Their needs will drive the industry, just as Gamers and overclockers, not
businesses, have driven PC development since 2005.

This is without considering growth in low-latency interactive services like geo-data and mapping/overlay applications: grown-up versions of Google Earth. "Augmented Reality" devices and games are already being sold - they take latency and bandwidth needs to whole new levels.

The on-going revolutions in use and demand caused by the iPhone in 2007 and iPad in 2010 should alert any planners, especially those doing networks, to the inevitable:
What you don't see will hurt you.
The only constant we've seen in the Internet since the web arrived in 1991 is exponential growth, akin to Moore's Law for CPU's. There is still no sign of it stopping, just easing down to a mere 12-month
doubling period.

No comments:

Post a Comment

Note: only a member of this blog may post a comment.