Solutions vendors may be ready to support the migration of applications into the 'cloud' model – but will groaning public networks be able to cope with all the added demand?
Just as two decades ago client‑server architecture hinged on local area networks (LANs), the success of cloud computing rests very much on the public wide area network (WAN). Both emerged partly because of advances in data communications, while exposing the acute reliance of application performance on that of the underlying delivery networks. In cloud’s case the stakes are higher: now critical data is also being entrusted to external agencies, and there may be less control over these until the intricacies of cloud contracting are resolved. It could take years.
The upshot is that enterprises still need to focus on the WAN as a major part of their cloud strategies, bringing together the four perspectives of security, performance, cost, and benefits. As security and performance issues are mostly solvable – at a price – cloud strategies tend to spin around the cost and benefits.
The question is whether benefits from cloud computing – reduced infrastructure management costs; stronger focus on direct-revenue generating activities; improved staff satisfaction; and greater customer loyalty – are worth the escalating costs involved with data communications.
These costs may be incurred through increased bandwidth, measures such as WAN optimisation and caching to counteract that increase. The more people who adopt cloud models, the more it will become compromised, as uptake places greater demands on the available public network bandwidth. There are ways round this – but, again, they come at a cost.
In global bandwidth terms the impact of cloud computing will be just a blip compared with the massive proliferation of video traffic generated largely by social networking and content-rich sites such as BBC’s iPlayer and YouTube, and the fast-growing range of Internet broadcast and on-demand TV entertainment services. Add to this the traffic from other sectors heading cloud-wards, such as education and public services, and the problem is thrown into relief.
The impact of cloud congestion on enterprises is going – at some point – to be felt acutely. One aspect of cloud is that, unlike some other trends in IT, the drive has come not from enterprises impatient for a more flexible, cost-effective way of delivering services, but from application vendors.
This has been pointed out by analysts such as Forrester in its report ‘You’re Not Ready for Internal Cloud’, published last summer. Like it or not, it says, “you need to embrace the fact that your developers and line-of-business leads aren’t waiting for you to figure this out”.
The report emphasises that the economic benefits are real and far too compelling for enterprises to indulge denial, even though as the title of the report suggests, they may not be prepared for it. Its focus is squarely on private clouds for enterprises.
It is worth clarifying the definition of ‘cloud’, which is somewhat of a buzzword term for a general imperative towards shared infrastructures and delivery of services almost as if they are utilities. As was pointed out in E&T’s ‘Cloud caveats’ feature (Vol 5 #12), in its essentials the cloud defines a ripening, so to speak, of technological features that have been around for years, rather than a new approach to online computing. There is a temptation to define cloud as utility computing in which all IT resources, including software, storage, and processing cycles, are piped-in to the place of need – like water or electricity, albeit with the option of paying by subscription, rather than just by metered usage levels.
Such a model has been rejected by many large enterprises for critical applications, such as managing customer bank accounts, on the grounds that security and privacy could be breached by allowing the transactions to extend beyond the protection of firewalls that sit athwart the LAN and WAN. Legislation in respect to the terms and conditions of customer contracts can also impinge here.
This has led to the notion of the ‘private cloud’, where the shared infrastructure is deployed by the enterprise itself, resolving some of the security issues because now all the data can still be embraced within a single firewall. Care must be taken to ensure that private clouds still deliver the perceived benefits of the cloud model, of which there are three according to Forrester’s definition.
First, the cloud services should be standardised such that every software component and infrastructure element is a commodity that can be delivered repeatedly. Second, resources must be available on a pay-per-use basis, or perhaps subscription, so that costs are entirely based on consumption rather than up-front purchase of infrastructure. Third, services must be capable of being provided on demand and turned on like a tap, which means deployment must be automated.
These are challenging to deploy within a private cloud, which is largely why so few enterprises have done it yet. In practice, private clouds are likely to be implemented via some hybrid model, more akin to traditional hosting (or even outsourcing).
At the same time, such clouds will make use of already‑proven technologies in order to share resources and commoditise infrastructure components.
The most likely model to emerge will be the hybrid cloud – or perhaps “virtual private cloud” sounds better – where individual enterprises are given what appears to be their own dedicated infrastructure, just as a virtual LAN simulates a local physical LAN across a shared network spanning multiple sites.
Under this model, a cloud service provider deploys multiple security firewalls across a shared infrastructure, giving each customer a set of dedicated resources while deriving the overall economies of scale of a large public cloud.
Issues of trust remain with such a model, as the enterprise is relying on an external provider just as with any hosted service. Indeed no matter what form of cloud is adopted, security remains a major concern; this is why providers of cloud technology like IBM have set up dedicated departments to help customers develop cloud security strategies.
The greatest challenge lies in determining how much to spend in order to achieve the benefits from whatever form of cloud is deployed, bringing the spotlight squarely back onto network capacity. As Forrester’s principal cloud-watcher analyst James Staten points out, given the demand for better customer and user experiences in the emerging fullblown multimedia world where video becomes integral to more and more applications, cost of bandwidth could be a constraining factor.
The message is clear – cache at all critical points of the cloud infrastructure including Web servers, browsers, and at the edge of the network. Browser caching reduces page load times, server caching reduces communication time within applications for processes such as database access, while edge caching cuts network latency resulting from the time taken for signals to traverse long distances across the Internet.
All fast-performing websites have in common extensive use of caching, and the same goes for hosted services. Caching itself is not a substitute for good network design – it is good network design (see box-out below). So says the Forrester report, and there is little with which to disagree save perhaps to emphasise that WAN optimisation should perhaps be given an equal status to caching rather than a subsidiary one.
Actually, the two go together, the objective being to reduce the transmission of data, particularly the chit-chat associated with many applications.