Network – the forgotten ingredient to cloud computing
The discussion on cloud computing is gravitating around very few hot topics. Some of these are Data Privacy, Security, Social Enterprise, Pubic vs. Private and Multi-Tenancy or Not. The latter one probably will die away quickly as it is just a result of Oracle’s latest labeling strategy. There is one huge topic though that is largely ignored though without it cloud computing would not be possible. The network which is used to connect to the cloud.
Surely there is the impression that network is not an issue anymore and that you are well-connected anywhere in the world even through mobile phones. That universal view is not true though.
I remember some years ago I was discussing SaaS with a representative from the UNHCR (the UN Refugee Agency). To them cloud indeed was appalling as it meant to get rid of the need of installing something on a device. Their local agencies do get a budget and procure the hardware locally. This poses a huge issue in a sense of client software deployment. So having established this as a good foundation for a cloud computing discussion we came across a major roadblock. With cloud computing as the backbone of their collaboration infrastructure there would be issues based on network availability and strength. The UNHCR often operates in areas with little or no continuous network. Actually establishing phone and network basics often covers a significant part of their budget. So offline capabilities as well the ability to work across low-bandwidth and unsteady connections were a major business imperative we were not able to meet with cloud computing.
This might sound like a special case but I have had so many customer meetings where one of the key questions was “We have this office in with an expensive and low-bandwidth network connection. Can cloud computing help?” And the answer is sometimes yes and sometimes no. You could improve network use e.g. by sharing a link rather than sending a document to each and every participant. But let’s face it that is not cloud computing specific. And if someone prints or downloads a 500MB presentation resulting in the network for the whole office being blocked, cloud computing is not the silver bullet. Only behavioral changes could solve that.
So the availability of a network at all, the cost of network and the user behavior do have a huge impact on the decision.
In case you are faced with situations as described above you need to start thinking hybrid and slightly left and right of mainstream architecture. Create solutions that rely on asynchrony communication and maybe allow you to build local hubs that only need regular uploads while keeping your local environment alive autonomic.
Network Performance / Network Bandwidth
Network performance is another area of concern as mentioned above already. One of Microsoft’s very first BPOS customers had huge concerns with regards to network performance and the resulting performance of their email system. They were used to have local satellite servers in all major locations around the world. These satellite servers handled the local traffic independently of the network connection to the rest of the enterprise. The cloud approach to serve all location from a central data center pair gave the customer the creeps. At that time Microsoft made a promise, remember it was one of their very first customers, to go for locally based servers as an extension to the cloud servers if needed. As a matter of fact it never came to that. The performance from the cloud service was not only sufficient but even much better than anything they had ever experienced before. So based on a good network cloud computing can deliver a fantastic experience. It only becomes difficult when it gets to the limits of the network. The cloud providers usually have a very good understanding of the network performance needed and some even have tools to calculate the bandwidth needed.
Once upon a time I did a project where we served a customer situated around Europe and in South Africa. We created a central system in Munich to serve all customer locations. The interface was built on servers running the OS Winframe (for the younger audience, Winframe was a separate OS based on Windows NT 3.5.1 and sold by Citrix. Only with NT4 the relationship changed and Metframe became a server extension for the Microsoft OS). The clever thing with Winframe and its protocol ICA was that you only transmitted the visual representation over the net and not the whole data. So in that sense it was perfectly suited for the task to connect through low-bandwidth connections. Anyhow we ran into an issue with the customer in South Africa which we had not anticipated. The latency time from Munich to South Africa was about 1s. One second might not sound much but created resistance. It confused the customer’s employees down there. The major issue indeed was the cursor and its representation on the remote screen. Anybody who tried to write an email or text document in the browser and lost the cursor has an idea about the issue. Also the blinking did not properly show due to the latency. We got the issue solved by working with Citrix to have the cursor created in the ICA client software rather than having that transmitted over the net but it gives you an idea what impact latency could create.
So where does latency matter nowadays. First of all it matters generally with cloud computing. Whether you have a roaming environment that could be in any data center around the world or you stick to fixed data centers latency could become an issue. Email is the least affected application but think of communications through voice and video or joint editing experiences. Latency cannot only become an issue it surely is something worth considering prior signing up for cloud services.
Service Level Agreements
You could have the best service level for the availability and performance of your cloud service in the world; it would be wasted if your network would not stand up to the same standard. When talking about service levels we talk about chains. And in a chain it is the weakest link that breaks first. The same applies to the chain of IT. In classic desktop outsourcing I tried to drive the providers into the direction to guarantee end to end SLAs or at least provide end to end monitoring. For cloud computing it is you as a customer that is accountable for the chain and each of its links.
So ensure that you do understand the service levels, how and where these are measured.
Four Mitigations Strategies to avoid Network Issues
So now that we have listed the issues it is time to look at mitigation strategies. These might sound basic but often there is no magic in IT excellence but just rigid execution of basics.
- 1. Analyze your needs
The most cloud computing providers do have a very good view on the network needs. They do have vested interest to share this information with you. Some even have built calculators based on application type, user profiles and usage scenarios. Search online and also ask for this input.
- 2. Choose the right network provider
Since most cloud computing providers do not support private lines into their data centers anymore you need to ensure that you choose your network provider according to your needs. Surely cost is a factor that plays a role in your decision but there is more to it. What is the SLA offered? Is the provider serving your cloud computing provider as well? If you choose the same provider for the network it increases quality and reduces risk of handover mistakes between networks.
- 3. Try out
The nice thing with cloud computing is that you can trial with little investment and little commitment. If there is an area of concern, test it. Make sure you have the right input to your decision matrix.
- 4. Keep an eye open for future developments
There is a persistent rumor that both Microsoft and Google do buy network capacities and create their own network globally. This could result in any of them offering a bundle of service and network. This would alter the game completely at least on the SLA side of things.
Talk to the service providers considered. Maybe a major change is just around the corner and they would welcome you as a pilot customer.