Gather ’round, folks – Grandpa has a story to tell. (OK, I’m not a grandpa and I don’t expect to be one any time soon, but as I’ve journeyed back in my memory to write this post, I sure feel old…)
For as long as it has been possible to have a full-time Internet in a residential home, I’ve been running my own home network, and I want to share its story: it’s The Little Network That Could, or TLNTC, as I sometimes call it.
(This will be the first in a series of posts covering design and implementation of a range of technologies in the Linux and networking space. I expect it to take me a while to finish. If you like this part of the story and want to read another network’s story while you’re waiting, check out Tom Eastep’s Shorewall documentation. Tom has done a great job over many years in providing detailed documentation of his configuration and design choices, and teaching a lot of people about networking in the process.)
Back in the bad old days (the late 1980s and early 1990s, in my case), those of us who were technical enough would connect to the Internet (which was only used for research and development purposes at the time) by a dial-up modem for long enough to download our emails and news spools into our offline text-based readers. This is because access was billed by the hour (or hours per month) at prices which meant that almost no one could have a full-time Internet connection – most connections were maintained by universities and commercial research labs. So my knowledge about networking was gained at work where I administered large commercial Unix systems, and mostly from the Unix host side (the network infrastructure was handled by another team).
In the early-to-mid 1990s, the Internet was opened for use by commercial organisations. This sparked a burst of growth which resulted in commercial ISPs providing more affordable Internet access (although it was still really only usable for people with reasonable technical skills), which eventually got to the point where it was affordable to have a full-time Internet connection over 56K modem. If my memory serves me correctly, this was around 1998, and I think my first full-time connection was via APANA.
I had been using Linux since around kernel version 0.97 (on the long-defunct SLS), and my gateway machine at the time when I first connected full-time was probably running Red Hat Linux 5.x (not to be confused with Red Hat Enterprise Linux 5.x, which was about 9 years later). I’m a little hazy on the details of this, although I do distinctly remember in 2000 deciding to upgrade from Pinstripe (6.9.5) to Guinness (7.0) without downloading the whole distribution first – it took 3 days of continuous downloading over my 56K modem, and failed in the middle twice, but eventually worked!
Around 2001-2002, Linux became well-enough accepted in the enterprise that I was able to shift my focus at work from what I considered to be increasingly unwieldy commercial Unix (and its overpriced proprietary hardware) to the much more frequently-updated and flexible Linux distributions – mostly Red Hat Enterprise Linux and SUSE Linux Enterprise Server in my work life, and Debian at home (Ubuntu was still a gleam in Mark Shuttleworth’s eye). Around the same time, Internet connectivity options were improving, and at home we switched from 56K dialup to cable modem and later to ADSL (which was actually a bit slower, but gave us a wider selection of ISPs).
Once our family had downstream connectivity in the order of megabits per second and upstream in the order of hundreds of kilobits per second, the benefits of local network infrastructure really showed themselves: we could have a local HTTP proxy (squid) and DNS cache/forwarder (BIND), which significantly improved web browsing performance (especially when multiple users were browsing), whilst still having enough bandwidth to receive and send email to and from our local server behind the scenes.
A change in my home network’s role came in late 2006, when I went into business for myself – what had been a fairly organic start became a rather more focused strategy, and the first part of my thinking around TLNTC was born. I was consulting on Linux, networking, and software development, and I dedicated around 15-20% of my time to research and professional development in order to keep my skills sharp. I needed more from my network than just a gateway for faster Internet and email access for my family – it had to be a proving ground for the technologies I was recommending, installing, troubleshooting, and maintaining for my clients. What was needed on site had to be demonstrated at home first, and workable over the long term.
Even though I’m not doing independent consulting at the moment, TLNTC still fills this role in my technical professional development, and this is why I’ve continued pondering its purpose and characteristics.
More than a home lab
A lot of engineers who work with Cisco, Juniper, Microsoft, or VMware products at work tend to have a bunch of gear at home, which they fire up when needed to study for a certification or build a proof-of-concept. Such home labs are often composed of retired enterprise equipment pulled out of a data centre rack or acquired on eBay from second hand vendors, although depending on budget and type of equipment, so more dedicated labbers might buy new gear. They often live in a full 42RU 19-inch rack in the garage, and are so noisy as to make other members of the household complain about jet engines and such when they are fired up. So their configurations can be short-lived and don’t necessarily need to be practical to maintain in the medium-to-long term.
I do use TLNTC as a test lab and I have some equipment that only gets turned on when I need it, but my focus is on applying learning to create a usable, reliable infrastructure rather than learning for its own sake. In short, it is designed to be an enterprise network in miniature. To that end, I’ve implemented a number of components which I generally encounter in enterprises, but would never recommend to home users unless they have similar goals:
- 802.1X authentication for wireless networks
- multiple layers of packet filtering firewalls, with OSPF and BGP for routing
- OpenVPN and IPsec VPNs
- IDS using full packet capture from a monitor port
- GPS-synced stratum 1 NTP server
- IPv6 through most of the network (more on this later in the series)
- URL filtering using a dedicated proxy server
- Network Monitoring System (LibreNMS) integrated with Opsgenie for alerting
Despite these similarities with enterprise networks, there are also differences:
- I strive as much as possible to only use well-maintained Free Software. I do run some proprietary software, including Junos on my main EX2200-C switch and Ruckus Unleashed for wireless, but these are the exception rather than the rule. When I first started consulting, this was sometimes a limitation, but it’s becoming less and less so. Nowadays I can usually find an Open Source technology for almost any enterprise software niche if I look hard enough.
- Performance and availability are generally lower priority for me than cost and noise. That’s not to say I don’t care about them at all, but there’s a balance to be struck. All my servers run RAID and some of those RAID sets are on SSD for speed, but generally I aim for small, quiet, and cheap. If I need more storage space, I’ll generally go for a spinning rust drive, due to the lower cost per TB. If a piece of server or network hardware dies, my family waits until I can get it repaired or replaced. If they urgently need Internet access, they use the hotspot on their phone. If they need email, they fall back to their gmail account temporarily.
- Core routing and firewalling happens in software rather than hardware. This is partially because VMs and containers are easy to modify and adapt, but also because firewall and router vendors have so consistently failed to produce platforms which are easily and frequently updated. I may take this point up in a later post in the series, but for now I’ll just say that I have found image-based software distribution such as that used by Cisco and Juniper much harder to manage and update than standard Linux distributions based on dpkg/apt or rpm/yum. Because of this, I don’t use dedicated firewall appliances, but build them from standard Linux distributions.
But it’s great for learning, too
I think there is also a learning benefit to taking the “mini-enterprise” approach to the home network: not only does the learning serve the infrastructure, but the process of implementing the infrastructure cements the learning. This means when I put technology on my resume, I do so knowing that I can confidently answer questions about it from experience rather than rote learning.
How mini is my mini-enterprise network?
To give an idea of scale, here’s a quick overview of what comprises TLNTC:
- 23 VMs or containers running across 3 dual-core VM/container hosts; 36 GB RAM total
- 3 small switches (all routing-capable, but running as L2 switches), total of 27 ports
- About 10 VLANs, each of which (for the most part) maps to an IPv4 and an IPv6 subnet and thence to a firewall zone
So clearly this is not a large network, but it’s considerably more complex than the average home network. Based on my experience during my time in consulting, it’s probably similar in size and complexity to the network of a small business with 25-100 employees, depending on how technical their work is.
Why not cloud?
A big question I’ve recently asked myself (and been asked many times, particularly when I tell people I run my own mail server) is: why aren’t you just putting this all in the cloud? Given that my day job involves working with AWS & Azure public clouds and container technologies like Docker and Kubernetes, I did seriously consider doing this, but decided against it on two grounds:
- I would still need most of the same on-premises hardware, and
I used the AWS and Azure pricing tools to work out how much my infrastructure would cost to run in their respective clouds. Azure’s pricing tool told me that my virtualised workloads would cost $60K to run in their cloud over 5 years, and $11K on-prem. AWS’ tool told me that I would save $273K over 5 years by moving my workload to their cloud. In reality, I’ve spent less than $7K on hardware in the past 10 years, and if I’m generous, maybe $5K on power over the same period.
Obviously this is not an apples-to-apples comparison since public clouds offer many features and services which my network doesn’t, but clearly if I don’t need all those services and I continue to prioritise cost over availability and performance, cloud is not the right answer. VMs and containers work pretty much the same on-prem as they do in cloud, so I’m not backing myself into a corner if one day I end up putting some of my home network’s workloads in public cloud. (This web site would likely be one of the prime candidates.)
[Edit: I couldn’t resist throwing this in – I just listened to IPv6 Buzz episode 055, where (around 49:30) Geoff Huston was heard to utter: “Folks who are running computers in their basement … are the dinosaurs of today’s age, and the enterprise networks that go with it are equally … dinosaur-reptilian-based bits of infrastructure.” I may circle around and respond to Geoff’s views in a future post, but in the meantime I only hope I can be a thoughtful, well-informed dinosaur. Triceratops is my favourite – can I be one of those?]
So that’s the beginning of the story of TLNTC – I hope it was informative. The next part of the story will be about TLNTC’s adventures in IPv6-land.
I’m happy to receive questions and comments in the section below, or via the usual social media channels (over on the right).