In our last newsletter, we shared a portion of a conversation with Thierry Grenot, CTO at Ipanema Technologies, concerning some of the details that one encounters implementing appliances to optimize the use of multiple transport technologies, such as MPLS and Internet-based services.
Today, we’re exploring some of the issues a bit more deeply with Thierry, especially with some of the more subtle capabilities that one may encounter.
Steve: Am I correct in my understanding that one of the primary benefits of hybrid networking is that both links may be used under normal circumstances for load sharing), but that all traffic will move to the other link(s) in the event of a link failure?
If so, can you tell us a bit about the metrics that are used for deciding when a link is “failing”?
In particular, must one totally lose connectivity, or is there a way that you also watch for degradation (especially in the case of Internet-based services) and start shifting traffic based on poor performance?
Thierry: The fundamental benefits of Hybrid Network Unification is that you always use (on a very granular and dynamic manner) the best network. Of course, if only one network is available, it has to be this one. But while networks are pretty reliable today, their end-to-end performance is very much fluctuating because of a) core network occupancy – for example Internet at peak time and b) usage of access lines by the enterprise applications. So Ipanema is selecting the best WAN based on application performance objectives, such as according to the effective performance that can be delivered by the user. The set of criterias depends on the application: voice will likely use the path with the least end-to-end delay between two points (here delay metrics are key) while file transfers will probably prefer the largest access (available bandwidth is the useful metric). With thousands of users and hundreds of applications, you can easily understand the combination of cases and all the variability that should be taken into account: manually impossible, a well designed optimization algorithm can make it.
Steve: Let’s chat for a moment about the interaction between using alternate links (hybrid networking) and network optimization.
In some cases, a well-optimized Internet-based link may approach the performance of an MPLS link.
How do you decide when to use which? Is this something that must be in some way specified (such as on a per-application basis), or does your appliance help do this – or at least provide “recommendations”?
Thierry: This question is actually related to the former one. It does happen that Internet is faster than MPLS, or that MPLS is less loaded than Internet. And so on. So the point is to not decide in advance: we don’t know tomorrow’s weather. We use a real-time sense and response principle that understand what is the performance for all networks and select the “best one” from a set of several criteria that depends on the application – what we call our ‘application performance objectives’. No need to be too fast on file transfer – no need to be too large for voice. What about video?
Steve: Most optimization solutions to date have been designed primarily for use between premises. However, we are now in the midst of a massive shift in some cases to cloud-based computing.
Can you give us a brief overview of how you can provide optimization for cloud services? (Of course, this is probably a sufficiently large question to merit its own discussion at some point.)
Thierry: I agree with you Steven. this is a fairly large point! In a nutshell, most of cloud applications are over the Internet (“THE” cloud)- This radically transforms the traffic matrix, from the traditional hub-and-spoke to a weird hub-and-hub configuration. Cloud computing –> Simplifying IT + Complexifying networking! A good start to talk about control and optimization, isn’t it?
And we’ll look forward to continuing this discussion, indeed. In the meantime, please feel free to join us to dig a bit more deeply into this discussion at Webtorials.
