The branch office situation may be considered from two main viewpoints: the network and the applications. Let's start from the latter:
- First of all, applications are not equal for the business. Some of them must absolutely work perfectly well, in any circumstances: this is generally the case of the business transactional applications (ERP…) and applications related to customers relationship. Some others may just be "good enough": while the delivery of an email with a delay of a few seconds will likely not rise problems, a customer waiting too long at the department store cashier desk or when organizing his next holidays trip at the travel agency, will not only impact the enterprise efficiency but also the customer satisfaction.
- Then, enterprise's network is falling prey of cannibals: studies (e.g. OVUM) show that in many cases +50% of the traffic is of recreational nature (Facebook, YouTube…) while only 15% are dedicated to business critical applications (we found 9% in our own study). I believe this trend will confirm with the ever increasing richness of contents (digital photos came from 2, then 5 and now 15 Mpixels) and the arrival of the famous Y generation in enterprise.
From the network point of view, let's just do the two following considerations:
- The connection(s) between branch offices and the [VPN + Cloud] network(s) is shared between all users and all applications, whether recreational or business oriented, either hosted in the enterprise datacenter or in the (private or public) Cloud. This common resource must then be properly controlled and allocated.
- Multi-homed branches (e.g. one access to an MPLS private network and another one to the Internet, or two accesses to the Internet) may benefit from a very efficient and particularly robust configuration (network load-sharing). In this case, VPN-hosted as well as Cloud-hosted application servers can be reached from each access, with an intelligent per-flow selection of the access ensuring the best performance and availability for the service, anytime.
To complete the description, we can add the highly dynamic nature of the traffic that varies with the very immediate users' activity. Traffic may be estimated 'on average' (e.g. at peak time) but cannot be predicted in real time: while traffic jam can be predicted according to time in the day, it is still impossible to forecast whether a car will suddenly appear right-hand at the near crossroad in the next few seconds. It is then necessary to ally medium-term planning with real-time driving.
…/…