Cloud location matters – latency, privacy, redundancy
CEO & Founder of Server Density.
Published on the 21st August, 2014.
This article was originally published on GigaOm.
Now that we’re seeing intense competition in the cloud infrastructure market, each of the vendors is looking for as many ways to differentiate itself as possible. Big wallets are required to build the infrastructure and picking the right locations to deploy that capital is becoming an important choice. Cloud vendors can be innovative on a product or technical level, but location is just as important — which geographies does your cloud vendor have data centers in and why does that matter?
Why is location important?
There are a number of reasons why a diverse range of locations is important:
- Redundancy: Compared to the chances of a server failure, whole data center outages are rare — but they can happen. In the case of power outages, software bugs or extreme weather, it’s important to be able to distribute your workloads across multiple, independent facilities. This is not just to get redundancy across data centers but also across geographies so you can avoid local issues like bad weather or electrical faults. You need data centers close enough to minimize latency but far enough to be separated by geography.
- Data protection: Different types of data have different locality requirements e.g. requiring personal data to remain within the EU.
- User latency: response times for the end user are very important in certain applications, so having data centers close to your users is important, and the ability to send traffic to different regions helps simplify this. CDNs can be used for some content but connectivity is often required to the source too.
Deploying data centers around the world is not cheap, and this is the area where the big cloud providers have an advantage. It is not just a case of equipping and staffing data centers — much of the innovation is coming from how efficient those facilities are. Whether that means using the local geography to make data centers green, or building your own power systems, this all contributes to driving down prices, which can only truly be done at scale.
How do the top providers perform?
The different providers all have the concept of regions or data centers within a specific geography. Usually, these are split into multiple regions so you can get redundancy within the region, but this is not sufficient for true redundancy because the whole region could fail, or there could be a local event like a storm. Therefore, counting true geographies is important:
Azure is in the lead with 12 regions followed by Softlayer (10), Amazon (8) and Rackspace (6). Google loses out, with only 3 regions.
Where is the investment going?
It’s somewhat surprising that Amazon has gone for so long with only a single region in Europe — although this may be about to change with evidence of a new region based in Germany. If you want redundancy then you really need at least 2 data centers nearby, otherwise latency will pose a problem. For example, replicating a production database between data centers will experience higher latency if you have to send data across the ocean (from the U.S. to Ireland, say). It’s much better to replicate between Ireland and Germany!
Softlayer is also pushing into other regions with the $1.2 billion investment it announced for new data centers in 2014. Recently it launched Hong Kong and London data centers, with more planned in North America (2), Europe (2), Brazil, UAE, India, China, Japan and Australia (2).
The major disappointment is Google. It’s spending a lot of money on infrastructure and actually have many more data centers worldwide than are part of Google Cloud – in USA (6), Europe (3) and Asia (2) – which would place it second behind Microsoft. Of course, Google is a fairly new entrant into the cloud market and most of its demand is going to be from products like search and Gmail, where consumer requirements will dominate. Given the speed at which it’s launching new features, I expect this to change soon if it’s really serious about competing with the others.
What about China?
I have specifically excluded China from the figures above but it is still an interesting case. The problem is that while connectivity inside China is very good (in some regions), crossing the border can add significant latency and packet loss. Microsoft and Amazon both have regions within China, but they require a separate account and you usually have to be based in China to apply. Softlayer has announced a data center in Shanghai, so it will be interesting to see whether it can connect their global private network with good throughput. As for Google, it publicly left China 4 years ago so it may never launch a region there.
It’s clear that location is going to be a competitive advantage, one where Microsoft currently holds first place but will lose it to Softlayer soon. Given the amount of money being invested, it will be interesting to see where cloud availability expands to next.