John Linton
One of the earler memories I have of IP bandwidh, apart from it being massively expensive in the mid 1990s (around the best part of $US1 million per year per TWO mbps) was the 'care with which it was managed. This was, of course, long before broad band and there was no such thing as downloading movies and TV shows and in those far off dial up days if you could get a flashless and practically graphicless web page to load in less than 20 seconds everyone was happy and didn't understand 'packet loss' and as there were no on line games there was no obsession with ping and trace route (which didn't exist and in fact the main users of those 'tools' today, teenage boys, didn't exist as internet users at all. So it was basically a technical buyer that rapidly spread with a total of something less than 50 mbps of IP connectivity between Australia and the rest of the world with Telstra owning over 90% of that connectivity and charging the few ISPs that existed at that time 30 cents PER MEGABYTE (not gigabyte) if they couldn't afford to buy direct from one of the very few US carriers that could provide trans-Pacific cable capacity (long before Southern Cross).
I have an early memory of one ISP I then worked for with their banks of 14.4 kbps modems and racks of inbound PSTN lines and the heat of the inadequately cooled room with the modems (against all manufacturers recommendations) stacked on top of each other dramatically increasing the heat dissipation problems and causing constant modem failures which would go undetected for days and often longer than that) before processes were developed to eliminate that horrible problem. But most of all I remember a much younger 'senior engineer' than he is today attempting to justify why the MRTG report of the day would not be a solid green oblong and enduring the scathing comments by the owner of the company about how he was "throwing money away" because there was some unused bandwidth between 3.30 am and 5.00 am (or whatever) on weekend mornings.
That was a long time ago, at least in terms of the cost and price of trans-pacific data transit costs. Today the cost of 2 mbps of trans-Pacific cable is, literally, one one thousandth of the cost it was 15 - 16 years ago but it is still an item that is 'managed' very, very carefully in many ISPs even at today's relatively very low costs. IP transit certainly is no longer the major cost of delivering internet services although, because of the continuing increase in end user customer downloads, quarter on quarter, it is still a significant cost. Managing IP transit capacity and its 'twin', customer connectivity capacity, remains the most important aspect of delivering continuously fast customer transit.
This has become much more difficult in at least one way in the 'broadband' era. With dial up there was never any contention on the customer's PSTN line which ws, effectively, a 'nailed down' copper circuit directly between the customer and the ISP - at least for the time the customer was connected to the service. Broadband only provides a PSTN (100% customer usage) line between the customer's residence and the exchange (and in the case of a RIM only to the RIM). From that point onwards the carrier 'aggregates' the traffic of all users onto its own singe (or perhaps redundant) back haul to the closest BRAS and then further aggregates the end user traffic between the BRAS and one or more State switches and then ......you get the picture. Finally the ISP only gets some semblance of control of service delivery when he buys the, again aggregated, bandwidth between the carrier's hand off point and the ISPs PoP/Switch in a State capital city.
So the customer and the ISP are entirely at the 'mercy' of the carrier in terms of relying on the carrier to provide sufficient back haul between the end user's exchange/RIM and the carrier's hand off point to the ISP - something that doesn't always happen but which the ISP, in the eyes of the customer, is totally responsible. Fair enough- as far as it goes - which is why every ISP duplicates the carrier's ts and cs in saying delivering any speed ADSL service is "subject to......" and is "not guaranteed in any way....." etc., etc. I suppose that's a very long winded way of saying that virtually no ISP that resells a carrier's service has any control over some significant aspects of the service which can only be solved by the ISP installing it's own DSLAMs and running and maintaining its own back hauls. Exetel doesn't fit in to that category.
The point that an ISP like Exetel does become 100% accountable for bandwidth adequacy is from the carrier's hand off point to and from the rest of the world. IP transit (and its associated caching) is the main component of whether the speeds a customer achieves at any specific time of day are adequate or otherwise.
In dial up days, before latency sensitive WOW's and Half Lives (I wonder whether the creators of that particular game had a great sense of irony when naming it?) and streaming movies 'off peak' time started at around 10 pm and lasted until around 9 - 10 am the following day. Only a very few people were very interested in sending and replying to emails between those times or doing any of the other, limited things, the internet was used for until relatively recently. So, with the exception of the ISP I referred to earlier who managed to use the 2 mbps of IP transit almost 100%, 100% of the time most MRTG reports would look like the silhouette a two humped camel with practically no usage between midnight and 8 am - a huge waste of a resource that cost so much money. In general terms, if you believed it would be possible to use 100% of the 2mbps 'pipe' a realistic ISP would barely get to use 50% of total theoretical capacity in the dial up days and it didn't get much better until broadband (if 256 kbps culd be called broadband) became available together with 'download management' software that - heavens to Betsy Paw! - this new fangled technical thingummy jig lets me set a download runnin' way after I'm in the land of Nod".
It took the US writers of such software a blink of the proverbial eye to promote their products to ISPs as a way of using the "wasted' IP transit capacity between 10.30 pm and 9.00 am and so "off peak" was born. Ever since 1995 I, and I assume a myriad other people responsible for the financial well being of an ISP have been attempting to replicate the 'green oblong' but (unlike the first person I knew who demanded this) using the totality of the bandwidth without ever causing packet loss because too much usage was saturating the available capacity at any point in time.
Exetel's "off peak" is currently 12 midnight to 12 noon and we have just introduced a process where any customer can select their off peak period for any given month from three different period (12 - 12, 1 - 1, and 2 - 2). This is the first step along the path leading to the 'Holy Grail" of bandwidth management a "green oblong MRTG report with zero packet loss". Our current utilisation is somewhere North of 85% and we believeit mmay be possible to increase that utilisation, without packet loss, to North of 90% before the end of the calendar year and then North of 95% by the end of calendar 2010. You might think that's a lot of effort to go to far a 10% saving of a cost that will continue to fall and you may well be correct in that view. But there are several other aspects of efficiency gain in IP transit utilisation that significantly add to the simple 10%gain (if in fact that could be achieved).
However the real issue is that if any commercial company sees an inefficiency in its operations and makes no attempt to correct it then the whole business will eventually be damaged by the most damaging of all attitudes in commercial life - "that's the way it's always been and there's nothing that can be done about it". In a technology business that has, more times than I can remember, been the beginning of the end for some one or some thing.