We see a clear line between primary and secondary cities based upon these attributes. The definition leads some to suggest the number of primary cities is between 55 and 75 but this is a small number when you consider there are 487 cities in the world with more than one million inhabitants.
There is a clear concentration in the US and in Western Europe. Asia and South America are sparsely provided for and Africa, Eastern Europe and the Middle East lack any primary cities.From a population perspective, only 15% of the world’s population is within 25ms (RTT) of a primary city.
So having established some of the basis for the problem let’s look at why its important. For this we need to move from real geography to Internet geography. We see this as three, concentric circles depending upon latency. Latency, as most of you know, is the delay that information experiences in getting from its source to its destination due to the speed of light and various pieces of electronics such as routers and switches that it must traverse. There are other causes of network delays, such as congestion but these can vary for a wide range of reasons so I won’t be considering that in any detail.As most flows of information on the Internet are made up of a request and a response, latency is normally measured based upon round trip time. This is in milliseconds and will be twice the time for information to make the trip in one direction.CLICK! The first circle is for applications that can operate effectively from one location in the world.CLICK! The second circle represents a significant portion of Internet traffic at present. For good performance content must generally be served from at least one location on each continent. The Limewire CDN used by Youtube is a good example.CLICK! The final circle is for applications that are the most interactive. You can see some examples on the slide. In this case generally content in every city is desirable. Few applications have problems with a 1000km distance from source to destination but unless interactive applications are ultimately hosted in every population centre we will end up with ‘haves’ and ‘have nots’ on the network.CLICK! This is why we need to look at the secondary city problem.
Sydney is a primary city and has over 130MW of operational multi-tenant data centre capacity. Melbourne has grown rapidly over the last four years to be heading into primary city territory. The interesting this is to then look at all of the other cities. We have a range of cities with a very low share of data centres while we have very large distances (which exacerbates the latency issue) and we are in the early stages of a bandwidth transformation (which exacerbates the latency issue).
We’ve done some research into this. We looked at population weighted latency across all Australian metropolitan areas, across a spread of all fibre networks. This enabled us to answer an couple of questions:CLICK: If you want to get the best average performance across the whole populationCLICK: Host in MelbourneCLICK: If you want to serve as many people as possible with a low latency (say <25ms) such as for simulation and online gamingCLICK: Host in AdelaideThe other interesting conclusion from this study is that many future applications will need to be hosted in multiple locations in order to provide a good experience for all users.
Data Centres have operated in a similar fashion for the last fifty years. Data centre economics work best at scale and facilities are built to a fixed standard of tier level and power density per rack or per m2. Building in secondary cities is a challenge as - data centre pricing tends to stabilise across the world due to the fact that some applications can be hosted anywhere (the outer ring in my earlier slide). As a result significantly higher returns can’t be achieved in secondary cities. - it is very hard to estimate the level of demand in a secondary city. Requirements are soaked up by out-of-date facilites, facilities in other cities and broom closets. - The growth in demand is also hard to estimate. Local economic conditions can also modify the take-up rate - Finally, data centres typically last twenty years. If the required standard for data centres changes (across any of redundancy, density, security, connectivity and so forth) then the market can move away from a stand-alone facility in a secondary city.
We took a modular design from Dell that stood out as it offered traditional data centre economics with the ability to deploy capital progressively. Obviously it is also extremely flexible. Efficiency is also extremely high with a range of clever design elements putting the PUE down to 1.2. This improves the economics further.All components of plant and office space are also modular.
Site selection options become incredibly diverse. This is a photograph of part of the former Mitsubishi car plant in Adelaide. This building is being converted into an 8MW Tier II/III/IV data centre using modular principals.
Tier 5 Networks' Peter Wildy
Data Centres in Secondary Cities
• Primary Cities
– Supply and Demand in some kind of balance
– Pricing set on this dynamic
– Continual construction
• Secondary Cities
– Below economic market share
– Data Centres typically extended beyond their useful
– No clear view of demand as capacity shifts to
How do customers decide?
• Factors used by customers to determine
data centre location:
– Latency (for some applications), i.e. close to
customers, well connected
– Close to Geeks
– Close to head office
– Away from hazards, low risk location
– Cheap power
this range, so
cities are no
Applications that can handle high
latency >500ms only need to be
hosted in one location in the world.
Applications that handle
moderate latency and work
well with caching such as
video need to be hosted in
every continent. <150ms
applications must be in
every city <30ms
that have a small
amount of content
or those that are not
include web based
and those with thick
clients that can
locally. It also
includes those that
respond well to CDN
are highly interactive
such as online
various forms of
Where is the best place?
Q: Where should I host my applications in
Australia to get the lowest population
Q: Where should I host my latency sensitive
applications to serve as many Australians as
possible from a single location?
So why doesn’t everyone do this?
• Commodity pricing
• Capacity risk
• Slow fill risk
• Locked-in standard risk
Is there a solution?
• A solution would need to incorporate
– Allowing capital to be invested progressively.
– Providing a flexible architecture so that customer’s needs for
density and tier level can be met as they evolve.
• It would also be desirable to have
– rapid deployment.
– high efficiency.