Upgrade to Cumulus Linux 2.5

In the world of computer hardware, it is traditionally customary to tailor product updates so that their appearance on the market coincides with the spring or autumn sales surge. But in the software world there is no such thing - here new versions and updates come evenly at any time of the year.
So, the company Cumulus Networks decided to submit an update of its operating system Cumulus Linux to version 2.5 just a couple of days ago, on January 19th.
Briefly recall that Cumulus Linux is one of the network operating systems designed to be installed on compatible switches with ONIE (Bare Metal Switch) boot environment . We have already written quite a lot about the capabilities of this OS, since we consider the concept of “BMS + open OS” to be a very promising direction in the development of the network infrastructure in terms of the cost of the final solution, compatibility and opportunities offered.

I wonder what appeared in the new version? Then welcome under cat.


First of all, we note that the company has approved design guidelines. They look very thoroughly and include both tips on the architectural construction of various solutions, as well as impressive listings of ready-made options. Currently three such manuals are available:

• on working with VMware vSphere
• on working in an openstack environment
• on building infrastructure for a project with Big Data (Hadoop)

')
Major innovations are mainly related to working with L2 - as the company itself says, they got a little excited with how ready everyone is to switch to L3, so they tried to integrate all the most popular functions into their OS. By the way, this very clearly demonstrates how the company's flexibility, its willingness to respond to user requests, and how much time it takes to enter new features.

Firstly, such a thing as MLAG (Multi-chassis Link AGgregation), which allows using the LACP protocol and the link between switches, to create active-active paired connections, has been definitively and irreversibly registered in Cumulus Linux, which is extremely useful both in terms of network security channels from failures, and in terms of improving performance. To understand the meaning and beauty of the solution, simply look at its scheme:

Those who wish to have deep details about the features of the work of MLAG are reminded about the existence of a site with very detailed documentation at Cumulus. For our part, we only note that MLAG is a great way to remove such a traditional single point of infrastructure failure as a TOR switch. Interestingly, unlike competitors, Cumulus not only does not consider MLAG to be a substitute for STP, but on the contrary, it strongly supports and strongly recommends not abandoning it, since it provides reliable detection of network failures.

The next item directly follows from the previous one and sounds like an improvement of the VRR (Virtual Router Redundancy) protocol. Now it supports networking with active-active configurations, that is, compatible with MLAG.

Next we have to change the driver responsible for the operation of the VLAN. The new version of the driver, which has become available with the transition to Linux kernel 3.9, allows you to build a single common bridge with the same common spanning tree, without separating separate VLAN subinterfaces from the physical ports. As a result, this led to a significant simplification of work and an increase in productivity when working in large-scale L2-environments with massive use of virtual networks. Details are correct in the documentation .

Continuing our listing is the addition of LACP bypass. Its function is to keep the 802.3ad link active and send data through it even in the absence of a LACP partner. A typical use of this feature is to enable PXE boot on a host that is not configured for LACP itself, but is connected via a link operating in 802.3ad mode.

In addition, a number of routing-related software changes occurred: the Quagga package was updated to the latest version 0.99.23.1, the main way to work with configurations was the ifupdown2 package, many improvements were introduced related to routing in environments with a large number of participants, BGP support appeared dynamic neighbors.
To this was added the expansion of opportunities for ensuring quality of service (QoS): the opportunity to pause a link, the ability to configure DoS and the store-and-forward mode appeared.

As a manufacturer of hardware, we cannot fail to note that the list of supported hardware in this release has not expanded. The reasons for this, it seems to us, are quite simple and are on the surface: the current list is enough to close all the requested ranges: 1G, 10G, 40G, both on PowerPC processors and on Intel processors, which further increases the flexibility to create solutions for specific customer requirements. And once everything is closed, it means you can switch to working with the next-generation chips, such as Broadcom Tomahawk, which are already on the way.