John Linton ...that is the very real question.
In the not too distant future the combined caching engines that Exetel has put in place over the past three years will deliver over 1 gbps of customer data in peak times. This is mainly from the Sydney located Akamai 'cluster' and the Sydney PIPE interconnect (which isn't true caching but for the purpose of calculating data costs it is treated by us in the same way). This doesn't include the up to 800 mbps delivered from the PeerApp P2P caching 'cluster' which we 'cripple' in peak times and only allow full access after midnight.
We have never used the PeerAPP for caching non-P2P data but it could be used for that very easily. Our view of caching has changed over time but it remains a complication that we have avoided over the past 2 years as we have moved the network from Sydney centricity for absolutely everything to a more dispersed 'architecture' - firstly activating additional PoPs in all mainland capitals and then progressively adding more 'local' interconnects and data paths.
Caching engines, either from PeerApp or one of several dozen other caching solution providers, continue to fall in price and, who knows, in the economic conditions that are beginning to be seen around the globe those trends are more likely to steepen than reverse over the coming months. As we near our 'major' anniversaries for our IP bandwidth contracts we need to make a decision on whether we will continue the policy of buying enough 'pure' IP bandwidth for our peak time (non-P2P) usage or take the cheaper options available from various caching concepts and hardware.
A lot depends on the $A - $US exchange rate/forward cover buying by the IP bandwidth providers as that will affect their own SX and Japan cable buy prices. The dire predictions of a sub 50 cent $A are not encouraging but, personally, I'm hopeful that won't happen (maybe that's just wishful thinking). We currently have, at peak times 1.9 gbps of 'pure' IP with Verizon and 600 mbps with Optus with up to another 1.2 gb plus being delivered from Akamai, PeerApp and PIPE and WAIX peering.
This is almost double the traffic we were delivering at this time last year - largely thanks to the increases from the caching/peering boxes/connections. However we have increased the 'pure' IP quite substantially over the last 12 months as well - partly to cater for the growth in the overall business and partly because of the lesser efficiencies of operating a 'distributed' network.
Our buy price for 'pure' IP isn't particularly good because even at over 2.4 gbps we aren't a "major" buyer of IP bandwidth and I suppose it also means we aren't the best price negotiators in the Australian communications business. As times get even tougher financially in Australia it is going to be difficult to continue to deliver the best value communications services at the IP prices we are paying today as well as cope with the ongoing increasing usage per customer that ADSL2 demands. We will either need to get much better pricing for 'pure' IP or use more caching in Sydney and deploy caching 'solutions' in the 'distributed' PoPs.
The capex for the distributed caching solutions hasn't been worked out in any detail and, because we will maintain our "no leasing" policy throughout 2009, may prove a barrier to full implementation - however I'm almost certain that we can convince our current caching suppliers to 'accommodate' our cash flow needs to overcome that if it turns out to be necessary. The appeal (financially) of caching is obvious - it can dramatically reduce IP costs and can also dramatically increase delivery speeds - particularly in a distributed network. The downside is that it can also produce some unhappy experiences if everything isn't continuously managed very carefully.
We have begun to get some 'indicative' IP prices for post 1/7/09 and they are 'hopeful' at this stage but if I had to make a decision today I think I'd move more rapidly on putting caching in to the distributed PoPs as the costs, even at the pricing being tentatively suggested, isn't commercially sensible for us to commit to with our current growth plans. It's a pity that the PIPE delivery dates remain so uncertain as it seems, at least to me from the little I know, that transit from PIPE is way more than a year away which is of no use to us.
We will run some trials over the next two or three months to see what actually can be achieved and if it proves positive then we will probably freeze our current 'pure' IP expenditures at something below the current levels and use caching to handle the growth in customer traffic demand.
It will be a challenging, (God, I must stop using that pathetic word) but very interesting, engineering evaluation.