One of the most exciting announcements from the last AWS re:Invent was the Elastic Network Adapter (ENA) Express functionality that uses the Scalable Reliable Datagram (SRD) protocol as the transport protocol for the overlay virtual networks. AWS claims ENA Express can push 25 Gbps over a single TCP flow and that SRD improves the tail latency (99.9 percentile) for high-throughput workloads by 85%.

Ignoring the “DPUs could change the network forever” blogosphere reactions (hint: they won’t), let’s see what could be happening behind the scenes and why SRD improves TCP throughput and tail latency.

The hype generated by the “VMware supports DPU offload” announcement already resulted in fascinating misunderstandings. Here’s what I got from a System Architect:

We are dealing with an interesting scenario where a customer had limited data center space, but applications demand more resources. We are evaluating whether we could offload ESXi processing to DPUs (Pensando) to use existing servers as bare-metal servers. Would it be a use case for DPU?

First of all, congratulations to whichever vendor marketer managed to put that guy in that state of mind. Well done, sir, well done. Now for a dose of reality.

One of the tiny details Open Networking preachers conveniently forget to mention is the tendency of open-source software to use a gazillion small packages from numerous independent sources to get the job done. Vendors selling commercial products (for example, Cumulus Linux) try their best to select the correct version of every package involved in their product; open-source projects could quickly end in dependency hell.

netlab tries to solve the dependency conundrum with well-defined installation scripts. We recommend you start with a brand new Ubuntu server (or VM) and follow the four lines of instructions¹. In that case, you usually get a working system unless something unexpected breaks behind the scenes, like what we experienced a few days ago.

Here’s some common-sense view on hard-coded rules versus machine learning in network operations by Mark Seery – quite often we can specify our response to an event as a simple set of rules, but if we want to identify deviation from “normal” behavior, machine learning might not be a bad idea.

For more details, watch the Event-Driven Network Automation part of Building Network Automation Solutions online course.

Did you ever wonder why everything in IT becomes slower over time? Our changed expectations and accumulated cruft definitely play a major role.. but it could also be hardware. For more details (and fun reading), explore Fail-Slow at Scale: Evidence of Hardware Performance Faults in Large Production Systems.

Did you like the traffic filtering in the age of IPv6 video by Christopher Werny? Time for part two: IPv6 traffic filtering details.

I keep getting questions along the lines of “is network automation practical/a reality?” with arguments like:

Many do not see a value and are OK with just a configuration manager such as Arista CVP (CloudVision Portal) and Cisco DNA.

Configuration consistently is a huge win regardless of how you implement it (it’s perfectly fine if the tools your vendor providers work for you). It prevents opportunistic consistency, as Antti Ristimäki succinctly explained:

I got this question from one of ipSpace.net subscribers:

My VP is not a fan of overlays and is determined to move away from our legacy implementation of OTV, VXLAN, and EVPN¹. We own and manage our optical network across all sites; however, it’s hard for me to picture a network design without overlays. He keeps asking why we need overlays when we own the optical network.

There are several reasons (apart from RFC 1925 Rule 6a) why you might want to add another layer of abstraction (that’s what overlay networks are in a nutshell) to your network.

It took vendors like Cisco years to start supporting routing protocols between MLAG-attached routers and a pair of switches in the MLAG cluster. That seems like a no-brainer scenario, so there must be some hidden complexities. Let’s figure out what they are.

We’ll use the familiar MLAG diagram, replacing one of the attached hosts with a router running a routing protocol with both members of the MLAG cluster (for example, R, S1, and S2 are OSPF neighbors).

One of the comments I received after publishing the Use VRFs for VXLAN-Enabled VLANs claimed that:

I’m firmly of the belief that VXLAN should be solely an access layer/edge technology and if you are running your routing protocols within the tunnel, you’ve already lost the plot.

That’s a pretty good guideline for typical data center fabric deployments, but VXLAN is just a tool that allows you to build multi-access Ethernet networks on top of IP infrastructure. You can use it to emulate E-LAN service or to build networks similar to what you can get with DMVPN (without any built-in security). Today we’ll use it to build a VRF Lite topology with two tenants (red and blue).

Cloudflare has been using ingress anycast (advertising the same set of prefixes from all data centers) for ages. Now they did a giant leap forward and implemented another “this thing can never work” technology: egress anycast. Servers from multiple data centers use source addresses from the prefix that’s advertised by all data centers.

Not only that, in the long-established tradition they described their implementation in enough details that someone determined enough could go and implement it (as opposed to the typical look how awesome our secret sauce is approach from Google).

Time for another netlab video: after explaining how netlab fits into the virtual lab orchestration picture, let’s answer the following question: what exactly can netlab do?

After VMware launched DPU-based acceleration for VMware NSX, marketing-focused websites frantically started discussing the benefits of DPUs. Although I’ve been writing about SmartNICs and DPUs for years, it’s time for another closer look at the emperor’s clothes.

What Is a DPU

DPU (Data Processing Unit) is a fancier name for a network adapter formerly known as SmartNIC – a server repackaged into an interface card form factor. We had them for decades (anyone remembers iSCSI offload adapters?)

A while ago, I wrote a blog post explaining why we should (mostly) disable ICMP redirects, triggering a series of comments discussing the root cause of ICMP redirects. A few of those blamed static routes, including:

Put another way, the presence or absence of ICMP Redirects is a red herring, usually pointing to architectural/design issues instead. In this example, using vPC Peer Gateway or, better yet, running a minimal IGP instead of relying on static routes eliminates ICMP Redirects from both the problem and solution spaces simultaneously.

Unfortunately, that’s not the case. You can get suboptimal routing that sometimes triggers ICMP redirects in well-designed networks running more than one routing protocol.

I planned to write a few interesting blog posts last week, but then got sucked into updating Azure Networking webinar. At least I got that completed 😊; the webinar materials now include these new Azure services:

• Azure Firewall
• Azure Private Link
• Azure Gateway Load Balancer

I also added descriptions of numerous new features: