<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Oren Zamir on Medium]]></title>
        <description><![CDATA[Stories by Oren Zamir on Medium]]></description>
        <link>https://medium.com/@orenzamir5?source=rss-13e6891b73f8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*2cdBmRuxMXoS3GeCT6K7QA.jpeg</url>
            <title>Stories by Oren Zamir on Medium</title>
            <link>https://medium.com/@orenzamir5?source=rss-13e6891b73f8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 09:42:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@orenzamir5/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[I Almost Replaced My Network Switch. I Built It a New Face Instead.]]></title>
            <link>https://medium.com/@orenzamir5/i-almost-replaced-my-network-switch-i-built-it-a-new-face-instead-1c4c3b337dd0?source=rss-13e6891b73f8------2</link>
            <guid isPermaLink="false">https://medium.com/p/1c4c3b337dd0</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[network-monitoring]]></category>
            <category><![CDATA[homelab]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[self-hosting]]></category>
            <dc:creator><![CDATA[Oren Zamir]]></dc:creator>
            <pubDate>Thu, 30 Jul 2026 13:40:51 GMT</pubDate>
            <atom:updated>2026-07-30T13:55:41.094Z</atom:updated>
            <content:encoded><![CDATA[<p>There is a NETGEAR GS724Tv3 in my home network rack. Twenty-four gigabit ports, no fan, quietly moving packets around for years without complaint. The hardware is fine. The hardware is, honestly, better than fine.</p><p>The web interface is from another era. Frames. A login page that forgets you. Counters that show you a number with no context, 4,182,993 packets, congratulations, is that good? To find out whether a port is having a bad day you have to write the number down, wait, come back, and subtract. There is no history. There is no graph. There is nothing that tells you something changed.</p><p>For a long time my plan was the obvious one: buy a newer switch. Something with a modern UI, maybe cloud management, a phone app. Two or three hundred dollars and the problem goes away.</p><p>I want to talk about why I didn’t do that, because I think the reason is more interesting than the project.</p><h3>The easy answer is usually the one that costs money</h3><p>Replacing the switch was never a technical decision. It was an admission that the only way to get a better interface was to pay a different company to write one for me. The switch itself was not the problem. The switch was doing everything I needed. I just couldn’t see it.</p><p>And here is the thing I kept coming back to: the data was already there. Every managed switch from the last twenty-five years speaks SNMP-a boring, well-documented protocol for asking a device about itself. Port status, byte counters, error counters, VLAN membership, connected neighbours, the MAC address table. My switch has been happily answering these questions the entire time. Nobody was asking.</p><p>So the real problem wasn’t a missing feature. It was a missing translator.</p><p>That reframing is only useful if you can act on it. Five years ago, “I’ll just write my own monitoring system” would have been a sentence I said out loud and then didn’t do-not because any single piece was beyond me, but because of what the pile looked like. How a Python async SNMP library handles retries. How to tell a counter that wrapped from a counter that reset. How to build a chart component that doesn’t lie about missing data. Each one is an evening. All of them together is a project that quietly dies at 30%.</p><p>That wall is a lot shorter now. Not gone-I’ll get to that, but shorter.</p><h3>What I actually built</h3><p><strong>It’s called netdash</strong>. FastAPI and SQLite on the backend, Next.js on the front, the whole thing in Docker Compose so it comes back up on its own after a power cut.</p><p>It polls the switch every thirty seconds and turns the answers into something a human can read:</p><p>A picture of the actual front panel-two banks of twelve, odd numbers on top, even below, with the top row’s connectors rotated 180° so the retention clips face outward, the way they really do. I built that from a photograph of my own unit. It sounds like a silly detail. It’s the difference between reading a table and recognising your switch.</p><p>Real traffic graphs, in bits per second, with history that survives restarts. Port names I chose-“QNAP”, “Office5”, “Music Room PC1”-instead of g16.</p><p>A health score per port that tells you not just that something is wrong but <em>what</em> deduction was applied and why. The MAC address table, so I can answer &quot;what is actually plugged into port 9&quot; without walking to the rack.</p><p>And alerts. If a link drops, if a port starts throwing errors, if a gigabit device negotiates down to 100 megabits, I get a message. Discord, in my case, but it also does Telegram, email, or a plain webhook you can point at anything. With quiet hours, because I do not need to know about a link flap at 3am unless it’s serious. And a weekly summary, because the useful question isn’t “is something broken right now”, it’s “is anything slowly getting worse.”</p><p>That last part is the bit I didn’t expect to care about and now care about most.</p><h3>The part nobody tells you about</h3><p>Here’s where I have to be honest, because the version of this story where AI writes the whole thing while you drink coffee is a lie, and telling it does everyone a disservice.</p><p>The code was the easy part. The hardware was not.</p><p>My first attempt to talk to the switch failed. Total silence, every request timing out-exactly what you’d see if the switch were unplugged. I checked cables. I checked firewall rules. I checked whether I’d fat-fingered the IP. The actual answer: this switch silently truncates SNMP community strings to sixteen characters. Its own web UI happily accepts a longer one, stores the first sixteen, and never mentions it. So the switch and I had different passwords and neither of us knew it.</p><p>I fixed that, and got a new failure that was worse, because it looked like success. The switch was now answering, but every query came back empty. Not an error. An empty result, from the very root of the tree. That turns out to be what a community string with no access rights looks like: accepted, authenticated, and shown a view of the world containing nothing.</p><p>Then the VLANs. There’s a standard table that tells you which ports are untagged members of a VLAN. Read it directly and my 24-port switch reported eighty untagged ports. The table is only meaningful when you intersect it with a <em>different</em> table-and no document says so; you find out because eighty is an absurd number.</p><p>Then LLDP, the protocol that tells you about neighbouring devices, which handed me a device name of è%±2X. Those fields aren&#39;t text. They&#39;re raw MAC addresses, and the switch declares them as text anyway.</p><p>None of these are things you reason your way to from a chair. Every one of them was found by pointing code at real hardware and watching it produce something stupid. That was the actual work, and it’s the work AI can’t do for you, because AI can’t see your rack. What it <em>can</em> do-and this is the leverage is collapse the distance between “I wonder if the untagged bitmap is the problem” and a script that tests it, from an evening down to about four minutes. When each hypothesis is nearly free, you check twenty instead of two. That’s not a small speedup. That’s the difference between finishing and giving up.</p><h3>One rule I refused to break</h3><p>Early on I made a decision that shaped everything after it: this system can only read. It can never write.</p><p>SNMP can absolutely reconfigure a switch. Shut down a port, change a VLAN. I wanted none of it. A dashboard I built in my spare time should not be capable of taking my house offline at 2am because of a bug I haven’t found yet.</p><p>So it isn’t a policy, it’s structural. The SNMP library is imported in exactly one file, which exposes only read operations. There’s a test that scans the entire codebase and fails the build if the write function ever appears anywhere. Port labels are stored in my own database and are never pushed to the switch.</p><p><strong>A monitoring tool that cannot break your network is worth more than one that can.</strong></p><h3>What it found in week one</h3><p>Port 17, “Music Room PC1”, was linked at 100 megabits-on a gigabit switch. It had presumably been that way for months. Nothing was broken, so nothing complained. That’s the signature of a bad cable or a bad port, and the only reason I know is that something was finally looking.</p><p>Port 12 reports one receive error per roughly 645,000 packets. Zero transmit errors, zero discards. That one-directional pattern points at the cable or the NIC on that machine, not the switch. It isn’t urgent. Faults like that don’t stay un-urgent.</p><p>And a broader one: while testing, I found the default public community string wide open on the switch with full read access. Anyone on my network could have inventoried the whole thing. I only found it because I was poking at SNMP in the first place.</p><p>None of these were emergencies. All of them were invisible.</p><h3>The point</h3><p>The tools available now don’t make you an expert in everything. I come from a DevOps background and networking isn’t new to me, but there were still plenty of details in SNMP, MIBs, and this particular hardware that I had never dealt with before. What these tools do is make those gaps much faster to cross.</p><p>The gap between “I have an idea” and “I have a working thing” used to be filled with skills you had to acquire first, in order, before you could start. That gap is much narrower now-narrow enough that the binding constraint isn’t the learning curve anymore. It’s whether you’re willing to look at a problem and ask if there’s a version of it you can solve yourself.</p><p>I had a switch with a bad interface. The obvious solution was to give someone money for a different switch. The interesting solution was to notice that the switch had been broadcasting everything I wanted to know for years, in a standard protocol, and that the only missing piece was something to listen.</p><p>It runs on a small Linux server in the same rack. It cost me evenings, not money. It does more than the switch I would have bought, because it does exactly what I want and nothing I don’t.</p><p>The switch is still the same switch. I just gave it a new face.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-HRx7K1vAGBp3NoEwUb91A.png" /><figcaption><em>NETGEAR GS724Tv3</em></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2t8Gx6HBDX0wXsEbeW8cHg.png" /><figcaption>netdash — Dashboard</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hr11DBlYFpykBC_oQcn7LQ.png" /><figcaption>netdash — Traffic</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GePx7F_Yi8q2Iwsw8dg07A.png" /><figcaption>netdash — Port Overview</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0MOCqQ052E-i9Ca8SFIP4w.png" /><figcaption>netdash — Port Graphs</figcaption></figure><p><em>netdash is a read-only SNMP dashboard for managed switches: FastAPI, SQLite, Next.js, Docker Compose. Built for a NETGEAR GS724Tv3, with a driver layer for others.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1c4c3b337dd0" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>