Regardless of how much progress we make in lowering WAN costs, this remains a major issue in the overall networking budget. And as we move toward cloud-based computing, the requisite amount of WAN bandwidth will only continue to increase.
As we've mentioned in several recent newsletters, there's a careful trade-off that must be considered. You can (relatively speaking) pay a lot for a high-performance service such as MPLS, or you can use an Internet-based service and save a lot of money but deal with a lower guaranteed level of service.
Consequently, there are companies such as Ipanema Technologies and Talari Networks that are creating products designed to empower enterprises to make the best of both worlds.
Recently, Steve had the opportunity to discuss some of the capabilities of Ipanema product with Thierry Grenot, CTO at Ipanema Technologies, and the discussion below is excerpted from that conversation.
Steve: Ipanema often discusses "hybrid networking" as a combination of MPLS and Internet-based VPNs. This makes a lot of sense because using these two services are at the extremes of high control, reliability, and price (MPLS) and minimal control, reliability and price.
However, there are other options available as well, such as WAN Ethernet services and even dedicated transmission services (such as T-3/E-3 and SONET/SDH). Can you please tell us a bit about which services are supported by your Hybrid Networking technology and how a user might go about deciding which services are optimal for their situation?
Thierry: Actually the transport technology does not matter. Hybrid networks can be made of any type of combinations of different networks. It happens that, as most of services are IP-based, in most cases we found MPLS and Internet combined in one 'unified network'. But we also see two MPLS, and why not Ethernet or pure virtual leased lines (like T-1 or T-3). At the end of the day, it is about IP traffic between clients and servers, that's all that matters.
Steve: I believe that Ipanema has the capability for appliances to use a variety of physical network interfaces and logical interfaces within the physical interfaces. Can you please explain how many physical interfaces are supported, and also how logical interfaces are separated within the physical interfaces?
Thierry: Currenty Ipanema proposes up to three virtual interfaces shared on a unique Ethernet physical interface (on the WAN side of the ip|engine appliance). Multiplexing is provided either through TOS/DSCP marking or destination MAC address selection (spoofing). We will also propose devices with multiple WAN-side Ethernet interfaces.
Steve: So, since there are two Ethernet interfaces, what is the most common place for the appliance to be placed? Between the LAN switch and the router? And then the router will choose the appropriate WAN interface(s)?
Also, since this placement could be viewed as a possible "single point of failure," do you have "fail to wire" capabilities?
Thierry: You want all the details, Steven, don't you? Concerning the appliance locations:
- Either there is only one router with two access lines – so we're mapping two 'virtual' interfaces on a single Ethernet line. Note that there is no need for 'virtual routers' (VRF) – just a static policy telling /if/ TOS bit x = Y /then/ go left /if not/ then go right.
- Or there are two routers (one per line) and here two options are offered:
- The same TOS field is used to ask the first router (usually the default gateway) to direct flows to the other one (usually through a line between them): /if/ TOS bit x = Y /then/ send flows to the other router /if not/ then take care by yourself
- Or MAC re-writing to directly aim at the appropriate router thanks to an intermediary switch.
There is no constraint about router, networks, bandwidth, etc.
This does not raise availability features (the Ipanema appliance would fail to wire in case of failure) but may require an additional device (low cost switch or hub). This is why we will provide a multiple WAN interface device very soon.
Steve: I assume you have the capabilities to scale to a meshed and/or partially meshed network. Can you give us a feel for the scalability of the solution, in terms of number of site supported, etc.
Thierry: Application traffic is more and more meshed (just think about voice traffic and even data center 'deconsolidation' caused by cloud computing and SaaS) so it is important to work in an any-to-any topology. The Ipanema system works well for very large networks having thousands of sites, with an actual average customer size of approximately 100 sites.
In the next newsletter we'll continue the conversation with Thierry. But, in the meantime, you can both get a step ahead and participate in the conversation with Thierry at Webtorials.