Midokura is releasing its Midokura Enterprise MidoNet (MEM) 5.0 platform this week, providing users with features that give organizations more visibility into Software Defined Networking (SDN) deployments.
MidoNet, Midokura's SDN platform, was open-sourced back in November 2014. Midokura CTO Pino de Candia explained that the new MEM 5.0 release is based on MidoNet (MN) 5.0, which was first released in October 2015. Midokura's product roadmap has MEM updates set to be released every 6 months, with MN updates every 3 months.
From a numbering perspective, de Candio noted that MEM and MN version numbers will now be kept in sync, so there will be a MEM 5.2 release in July, MEM 5.4 in January 2017, and so on. Prior to MEM 5.0, the numbering was not in sync; the last MEM version prior to the new 5.0 release was numbered version 1.9.
Of the MN 5.0 release, de Candia said that two examples of added innovations are service chaining and port mirroring. Building on top of that, the commercial MEM 5.0 product adds MEM Insights, providing visibility into flow history (including the detailed logical topology path of each flow), traffic aggregation reports, and traffic counters and aggregated traffic counters.
The prior release of MEM (version 1.9) did not have Insights but did have the MEM Manager user interface, which enables administrators to view and manage logical topology.
Also key to MEM 5.0 are high availability and fault tolerance, which are achieved by way of multiple mechanisms, including the use of stateful Layer 4 services. Those services include source NAT, Load Balancing as a Service, Firewall as a Service, as well as security groups and port-level firewalls, all with distributed data plane implementations.
"This is different from having multiple and/or scalable sets of instances through which traffic must be routed. Every MN Agent is able to locally act as a slice of the Stateful L4 service," de Candia explained. "MN achieves this fault tolerance by intelligently distributing the state of these devices and sending other Agents notifications of state changes as needed."
Going a step further, de Candia explained that other control plane requests are answered locally (without leaving the hypervisor/host): ARP replies from Router, ICMP errors (for reject routes and endpoint unreachability), DHCP offers and acks. Similarly, MN can locally proxy Metadata requests from VMs to the Nova/Metadata service available on the underlay.
"In this way, MN removes all the Neutron Agents and namespaces that run on the Controller node," de Candia said. "Since MN uses its own horizontally scalable L3 Gateway nodes (distributed across different racks) for N-S traffic, we have completely eliminated the need for, and bottleneck of, Neutron controller nodes in MN installations."
Looking forward, the next MEM release is version 5.2, currently scheduled to debut in July. Among the features that de Candia expects to be included are: Kubernetes and vSphere integration. Additionally, MEM Insights will likely benefit from integration with physical switches, starting with Cumulus Linux and other platforms in later releases. Another big item on the release roadmap is support for multi-site workloads.
"Specifically, the ability to create private links on which virtual routers in different sites and deployments can exchange traffic without crossing the private/public boundary; also, multi-site security mechanisms," de Candia said.
Sean Michael Kerner is a senior editor at EntepriseNetworkingPlanet and InternetNews.com. Follow him on Twitter @TechJournalist.