Category: Blog

I recently had the chance to work with a UniFi UDB-Switch during a network deployment. Long story short: It is an unmanaged switch with a 10 Gbit RJ45 uplink port and 7x 2.5 Gbit RJ45 ports. Basically your normal switch.

Where the UDB-Switch differs from the Flex 2.5 or other UniFi products is the fact that it has an integrated WiFi 7 uplink, meaning that you can deploy it without having to use any of the RJ45 ports to connect it to the rest of the network.

For the deployment I was part of, drilling through the walls or running cables was simply not an option. Thankfully, the distance we had to cover was short, so we contemplated whether or not the UDB-Switch could be the answer. We took the plunge and deployed it – and it worked beautifully for the simple setup of bridging a handful of wired devices across two rooms.

In fact, I was so flabbergasted by the device that I had to grab one for myself to play around with.

Unfortunately, while it is great as a device bridge, it falls short in several other key areas. I was not able to test this during deployment, so the following points are purely based on my personal tests with my own device.

No MLO

Yes, you read it correctly. The UDB-Switch, a device without 2.4 GHz support, lacks MLO. To be fair, it is supposed to come in a future firmware update, however keep in mind that the device has been on the market for about 6 months now – so a key feature that can be pivotal for enhanced performance is just missing from the product you can buy today.

On the 6 GHz band I was able to get around 1.3 Gbit/s in my iperf3 tests across my 10 Gbit network directly from my Gateway Fiber to the UDB-Switch, however keep in mind that these values mean next to nothing.

No LACP

I like the fact that the switch has 2.5 Gbit ports. However, some of my older multi port devices only do 1 Gbit per port; on other switches they benefit greatly from combining 2 or 4 ports into an LACP trunk (at least when multiple devices access them).

The UDB-Switch does not support LACP or bonding. This is not a case of having to uplink it via cable to enable the feature – it just doesn’t support it. Similar to the Flex 2.5 this seems to be fully intended.

Personally, I think this is a shame.

Media Disconnects and Uplink Changes

Switching between a wired uplink and a wireless uplink requires a complete reboot of the switch. This is somewhat baffling, especially if people purchase the device with the intent to have an optional wireless uplink in case the wired one goes down.

I was unable to determine whether this is a bug in the firmware or intended behaviour.

It runs HOT.

Similar to other Ubiquity UniFi devices, the UDB-Switch runs fairly hot, with idle temperatures around 60°C in my case. A bit of sleuthing shows that people report temperatures up to 80°C.

Thoughts

Ultimately a lot of my complaints can be waived away with the statement “that is not what the UDB-Switch is intended for”. And there may be truth to that.

However, the lackluster and outright unfinished firmware suggests a rushed product release. MLO STR is at least on the horizon, with signs of near-future release. However, as for my other points, I doubt any of them will ever get addressed.

The device is great if you need to bridge a gap and can easily replace a dedicated WiFi uplink client. Just be aware of the potential gotchas. My recommendation is to check the specs of the Flex 2.5, as the device seems to be basically the same (minus the WiFi uplink).

For more sophisticated switching requirements, you might fare better with a more potent managed switch and a dedicated WiFi uplink device.

If all you want is reasonable bandwidth and ethernet from, pardon the pun, the aether – then the UDB-Switch is a great choice. The people we deployed the UDB-Switch to are happy, bandwidth is not a problem (even during more congested scenarios).

Most important of all: No cables had to be routed and no holes drilled. And sometimes, that is all that matters. However strange that may sound.

Small oddity about Windows 10: When trying to shut down the computer, it started doing a very quick reboot (without going through the entire BIOS/UEFI/bootloader sequence) back to the login screen. The event log does not help. So what gives?

Turns out, there is a setting called “Fast Startup” in the power options:

Disabling the “Turn on fast startup (recommended)” option fixes Windows’ unwillingness to shut down.

I have no explanation as for why this happens, but at least it enables me to properly shut down my machine instead of choking my power button for 5 seconds to force a power off.

After a good long while, I have updated my local installation of LINQPad, a .net programmer scratchpad application that allows for easy computation and data analysis.

Much to my bemusement the application now features integration of LLMs like OpenAI or Anthropic. However, LINQPad also supports the use of a custom endpoint, meaning you can roll an Ollama setup and waste your own power instead of paying to waste someone else’s.

To configure Ollama, open the AI Settings and add a new AI provider.

When prompted to select the provider type, select “Custom OpenAI”.

Give the endpoint a proper name (i.e. Ollama) and set your endpoint to:

http://localhost:11434/v1/chat/completions/

Uncheck the existing OpenAI and Anthropic models and add your own local ones that you have installed with ollama pull <modelname>.

I found qwen3 to be a somewhat competent test model.

Using this setup, I was able to play around with the LLM integration into LINQPad. Unfortunately, as of version 9.6.6, there is no way to completely hide all traces of the AI stuff. If you are not interested in using these features, be it for monetary or compliance reasons, or out of conviction, you will constantly be reminded that they exist.

Right now I am in the process of migrating everything except telephony off my FRITZ!Box. After a lot of experimentation, sleepless nights and much frustration, I have come to the following conclusions about FRITZ!’s little all-in-one device.

Running the FRITZ!Box in IP-Client mode is basically impossible due to the sheer number of bugs. In this mode the router becomes just another client device in the network.

In theory this should be exactly what I want. It receives it’s IP address via DHCP and only acts as a SIP gateway. Or so you think. Because it doesn’t do jack shit. Once in IP-Client mode, the box no longer cares about WAN IP changes. It doesn’t even detect pulling all ethernet cables and just goes on being “online”.

Because of this, WAN IP changes do not cause a SIP REGISTER packet to be sent, meaning that you will have to wait the allotted time defined as REGISTER renewal in the original REGISTER response. In the meantime you cannot accept or initiate calls. Good stuff, FRITZ!

Now in the olden times people were able to activate telnet access to their FRITZ!Box and put a simple shell script into the debug.cfg to check for WAN IP changes and to reload the voipd if necessary.

But no worries, this feature has long been deactivated, so that absolutely no means of fixing a 15 year old issue are present anymore. Yes, you can restart the device through TR-064 via a shell script from another system, however I am not a fan of restarting an entire device because one teeny tiny option is missing. So… Good stuff, FRITZ!

And just to be extra annoying the FRITZ!Box decided to enable PVC on its own. Even finding the source of this issue took a good amount of time because why would I ever look at settings that I never changed to begin with? Good stuff, FRITZ!

While we are on the topic of settings: You cannot change the local domain name from fritz.box to something compliant with a .internal domain. Good stuff, FRITZ!, please ensure your fancy domain name keeps getting renewed.

All of these issues point to poor software quality or malicious compliance. Even without the ability to access the internals of the device and just by using the web interface (after several factory resets, to boot!) I was able to produce a ton of bugs and seemingly configuration corruptions when using the device in IP-Client mode.

Why telephony in IP-Client mode doesn’t regularly poll its own WAN IP to re-issue a REGISTER on change is beyond me. Why upstream gateways are not being accepted unless a static IP address is provided is also beyond me. Why this device is the only device on my entire network to behave so catastrophically uncompliant? I wish I knew.

There is no doubt that for regular households a FRITZ!Box is an excellent device. TR-069 automatically pulls the configuration for subscribers, WiFi configuration is dead simple, telephony has all the basic features you need. And all of that is pretty much working great when in router mode. But as soon as you switch into IP-Client mode, even after a full factory reset, the device breaks apart.

Skimming over 10-15 years worth of forum posts in search of solutions to seemingly wide-spread problems leaves me with the following conclusion: Instead of addressing long-standing problems within the networking core of the product, FRITZ! seemingly focuses on FRITZ!nas (because everyone and their dog uses their underpowered router as a NAS), on increasing the filter lists from 10 to 12 or on releasing another model of a FRITZ!Box with 1G ports.

The final verdict is that I will have to use the FRITZ!Box as a modem/router combo just so that I can keep the telephony features working reliably. Unfortunately it needs to sit at the front door because it cannot comprehend what a window is. Because unless it runs as a router, a lot of features of the FRITZ!Box simply stop working reliably or actively break.

Everything (including WiFi) apart from telephony will move behind the UniFi Gateway Fiber, using dedicated hardware.

What a shame, FRITZ! always stood for “not fancy, but reliable and compliant”. I’ve been using their hardware since the late 1990s, starting with the FRITZ! PCI ISDN card. Lately, their hardware offerings have been rather weak (why doesn’t the FRITZ!Box 5690 Pro have 10Gbit WAN ports? Why is there no XGPON/XGSPON support present?) and the reliability of the routers have been middling.

My 5690 Pro keeps crashing (as in: the device has to reboot) when certain WiFi settings are active. Extremely “Pro”. Makes you wonder if it will end up like a 7590 with hardware defects after 5 years due to maliciously misused components. Because we all love planned obsolescence!

While I do have my gripes with UniFi as well, it is at least “the Apple of networking stuff”: Nice user experience for dummies like me who want more control but are not Cisco-certified network gods. And I can change the gosh darn DHCP parameters like network boot and TFTP! Imagine that! And thanks to the SFP+ slot on the Gateway Fiber and a nifty X-GPON SFP module, I can quickly (!) bring up a backup GPON connection should the FRITZ!Box ever shit the bed, since all of the network topology is already behind (or will soon be behind) the gateway. Imagine that!

I purchased a Luleey LL-XS2510 a while ago. This SFP module acts as a fiber modem in GPON-based networks and has some creature comforts like configurable GPON serial numbers, vendor ID or MAC – essentially allowing you to “clone” your existing fiber modem to have a failover backup device ready, should your primary modem malfunction.

I wanted to use the module in a UniFi Gateway Fiber, simply to see whether or not the hype around Ubiquity UniFi was justified or not.

However, after configuring the Luleey with a media converter and pushing it into its own subnet, I came across a silly issue: I could not longer access the SFP module’s built-in web UI.

So, what is the situation?

• The Luleey SFP module has a static IP address of 192.168.10.1
• The default network of the UniFi Gateway Fiber is 192.168.1.0/24

Clearly, we need some routing here. The following instructions should get you up and running.

Note: This guide assumes that the port used for the Luleey SFP module is initially set to a LAN port. Please make sure it is not set to WAN!

Step 1: Create a new network

Go to Settings – Networks and create a new Virtual Network.

Set “Gateway Fiber” as the router and the zone to “Internal”.

Leave “Auto Internet Source IP / NAT” ticked but remove the checkbox for “Auto-Scale Network”.

Set the IPv4 address to 192.168.10.254 and the netmask to 24.

Set the VLAN ID to an unused VLAN.

Leave all checkboxes except mDNS unticked and set DHCP Mode to “None”.

Since the Luleey module does not seem to have IPv6 support, set the IPv6 Interface Type to “None” as well.

Save the network.

Step 2: Assign the network

Now open your port overview, select the port with the SFP module and set the native VLAN / Network to the network created during step 1.

Step 3: Create Policy Engine rules

Open System – Policy Engine – Policy Tables and create a new NAT Policy named “Translate Traffic to Luleey”.

Set the Type to “Masquerade” and the Interface/VPN Tunnel to the network created during step 1.

Keep the Source and Port to “Any”.

Set Destination to “Network” and select the network created during step 1. Keep the Port to “Any”.

Save the policy.

Step 4: Test it

You should now be able to open the Luleey web interface by opening http://192.168.10.1.

Caveats

As mentioned before starting, this only works for LAN ports. As soon as you assign the port to a WAN interface, the policy will no longer work.

However, unassigning a WAN interface from the port immediately restores access.

This is quite handy if you ran tests and want to check PON parameters or statistics about HEC errors. Since the module does not reset, the existing values from your prior tests will still be visible – giving you some indications about the quality of the connection.

Of course, console access works as well.

UniFi Caveats

There is one big UniFi-specific caveat: Just plugging in the SFP module will not bring it online in UniFi Gateway. Instead, the port will report failure due to signal problems.

To solve this, please connect a live fiber connection to the module. UniFi should stop throwing fits afterwards and allow operation of the module.

When trying to backup a Synology NAS, one has a few options. Synology’s Hyper Backup is great if you have relatively small shares or want to keep your NAS’ configuration safe. However, with larger volume or share sizes Hyper Backup becomes pretty unwieldy because it needs to finish a backup in one go, with no option to resume (i.e. after a power outage). This is bad because all of a sudden Hyper Backup not only has to roll back its already backed-up data, it starts from the very beginning again.

rsync has the ability to resume file transfers, however it is single threaded and relatively slow with an approximate 30-35 MB/s. This sort of transfer speed does not work for larger volumes or shares.

The solution I ultimately landed on was a two-step approach:

1. Mount your large share(s) via NFS on the destination NAS.
2. Install Rclone from the SynoCommunity community packages.
3. Do a normal local folder-to-folder sync.

rclone sync nfs_source_folder local_destination_folder --metadata --transfers 4 --checkers 16 --progress

With rclone, I get a nice and constant 98 MB/s with the ability to cancel and resume at any time.

You might be tempted to use SFTP instead, however this approach did not give me performance numbers even close to NFS.

Hyper Backups deduplication is nice, however I prefer having my backups finish within this decade over saving a few gigabyte.

The 14th of October 2025 is the day when Microsoft declares Windows 10 dead. The company expects regular home users to have migrated to Windows 11 by then.

Many users, myself included, are reluctant to switch over. The many horrifying news articles, often speaking of bugs so severe that only a restore from a backup will help, are not exactly inspiring confidence in Microsoft’s operating system.

Others will point to Microsoft’s statement that Windows 10 was supposed to be the last Windows version (according to Microsoft).

Personally, I simply cannot get used to the dumbed-down user interface of Windows 11. I need the backwards-compatibility of Windows Explorer, especially in regards to the Windows taskbar.

Windows 11’s new interface has been rewritten and redesigned, disregarding 30 years of backwards compatibility. I have no doubts that “you are just backwards” or “this is the future, old man” will be the comments. Unfortunately, these comments do not address the underlying issue.

The new user interface is not only breaking backwards-compatibility, it is also slow and impractical for power users, missing some basic quality-of-life features like being able to reposition the taskbar to another side of the screen.

Again, the Microsoft sycophants will crawl out of the woodwork and proclaim that all of these issues can be addressed with enough registry tweaking or third-party tools – missing the point once again.

I do not want to use a third-party shell to be able to use a program that I’ve been using for 25 years. I do not want to use a third-party hack (that breaks regularly after Windows updates) to restore functionality that never should have been cut to begin with.

Yes, ExplorerPatcher is a great tool. I’ve given it an extensive go in a virtual machine to see whether or not it makes Windows 11 bearable. It does, because it effectively turns Windows 11’s shell back into Windows 10’s shell. The more pressing questions are: Why do I need to use a tool to begin with and how long until Microsoft renders some of the tweaks to the start menu and taskbar (that I consider vital) completely unusable.

Personally, I do not foresee myself upgrading to Windows 11 anytime soon. No matter how great the under-the-hood changes of Windows 11 are, the user interface makes it a complete no-go for me.

I had hopes that Microsoft would address these long-standing issues over the years. And to be fair, some of them have been addressed – at least a teeny tiny bit. But power-users with the need to use some legacy software are trapped on Windows 10.

And what did Microsoft achieve, I ask you. A worse version of the same user interface we have had for the last 30 years. A taskbar that now needs to load for a while before even appearing. A start menu that is impractical to use. A new file manager that… well, at least Windows Explorer has tabs now. Good on you, Microsoft, you did something good. But you screwed up the context-menu, so you still won’t get a pat on the back from me.

This will become a thorn in Microsoft’s side over time. If you melt away many of the conveniences and third-party applications that anchored people into your ecosystem, while simultaneously antagonizing your old userbase with advertisements, dumbing down the operating system, an overall bad performance when it comes to the quality of updates and unwanted user metrics/AI integration, there is a real risk that people will just stay on outdated systems.

For me personally, switching to Linux is not an option. I came from Linux after wrestling with the open-source desktop for close to a decade. I have had nothing but issues with Wayland during my brief tests every few months, and last time I checked, two or three additional layers of audio subsystem abstraction have been added. I do not have the strength to migrate back and fight the same battles I’ve had in the early 2000s.

I also prefer my software to be written by people that are at least accountable – not by an organization that hires a shaman or a group that gets Trump Derangement Syndrome and purges people they deem “politically evil”.

Why can’t we just have the same desktop we had for 30 years? I cannot comprehend how that is too much of an ask.

For a while I have had an annoying issue with Citrix Workspace 2503.2. In the past, the Receiver application would hold and refresh my token, allowing for prolonged operation. I was able to show the main window over the course of the day and launch multiple applications.

At some point, I started receiving the error message “Unable to complete request” in the main window. The tray context menu switched from the standard menu with Sign In/Out, Connection Center etc. to a simple Open and Exit menu.

The Event Log showed that Receiver.exe was crashing due to this .net exception: The process was terminated due to an internal error in the .NET Runtime at IP 0x7395633A (0x73950000) with exit code 0x80131506.

So, what is going on? Simply put: The current version of .net Framework 8 (at the time of writing: 8.0.16) is causing some sort of regression that ultimately crashes background processes. This leads to the degraded user experience and service interruptions.

Uninstalling .net Framework 8.0.16 and re-installing 8.0.14 solves this issue. Be sure to install the x86 runtime for Citrix Workspace.

If you ever come across the EDC part of YouTube, you have probably seen the Traveler’s Notebook and it’s smaller sibling in passport size. A lot of people seem to treat it as a background for collages, an art project for Instagram or a gloried sticker album.

To me, these notebooks replace more traditional notebooks and planners. The advantage of Traveler’s refill system is that you can easily swap out parts of the notebook and add refills when needed, making it a more or less infinite binder system. Thanks to the various categories of inserts, I can easily add pouches to store keys or cards, small binders for documents and various types of paper.

I feel that it is important to have a completely analogue medium. I carry my passport-sized notebook with me whenever I leave the house, which gives me a quick notepad if I ever need to jot something down quickly – be it dates, prices, names or whatnot.

Smartphones are cumbersome to use and usually fail you due to poor network quality or a dying charge. Ink on paper is much less trouble and my handwriting easily outpaces my ability to tap away on my phone. My beloved Zebra F-701 snuggly hides away in the notebook, so I never have to worry about whether I have a pen on me or not.

When it comes to my work, I prefer the regular sized notebook. It allows me to quickly draft my ideas, so they are easier to bring into a clean form (i.e. when formulating requirements). The regular sized notebook clips two Lamy Safari fountain pens, each with Fine nibs and different colour inks. This helps me highlight, draw and correct information. I use dotted paper and clip it down, so I can work with both pages of a fold at once to award me more space.

I can only caution you to have an analogue medium as a workbook. Despite being a huge fan of computers for roughly 30 years now, I feel that there is no replacement for a good old sheet of paper and some imperfect scribbles.

A word of advise: Do not lose yourself in the perfection of formatting or trying to be smart about your layout. Instead, focus on what really counts: Information.

Jot down stuff. As soon as it is done, cross it out or mark it as done. Do not make your notebook your art project, make it something you actively use instead.

I am using two Chromium-based browsers on my system: Wavebox and Vivaldi. Vivaldi is my everyday browser and Wavebox takes the place of dozens of Electron-based applications, like Discord or Google Chat.

Recently, my CPU usage would spike as soon as both browsers were running. Opening the Task Manager in both applications revealed a CPU usage of 50% for a process called “Utility: Video Capture”. This only happens when both browsers are open, leading me to believe that this is a bug in Chromium itself.

I don’t need nor want video capture on Vivaldi.

On Windows, your best bet to disable the process is to create a new registry key that disables the feature altogether. Trying to mess around with flags within the browser did not yield any results.

I will use Vivaldi as an example here, however a similar procedure should be applicable to most Chromium-based browsers.

Open regedit and navigate to HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Vivaldi.

If the key does not exist yet, create it. Within the “Vivaldi” key, create a DWORD32 named “VideoCaptureAllowed” and set its value to 0.

After restarting Vivaldi, the process no longer burns through my CPU cycles.

I have been happily using the AKG K702 reference headphones for a few years and can fully recommend them – especially for the price. However, there are two small maintenance related things, that I want to highlight using this platform:

Once you have to replace the velour ear pads, go with third-party leather ones. Bought a set of Geekria pads from Amazon and have been pretty happy with them. Paying around 40 EUR for a set of official velour replacements from AKG that I personally find uncomfortable is hard to justify when there are cheap leather ones available for the price of one official pad. The leather ones can easily be wiped, making them a superior choice, in my opinion.

The other pain point are the elastic straps on the headphones. The design is pretty elaborate and accessing the elastics is a bit of a pain. Do not use a tack to try and get the grills removed, as one often reads on the internet. If you have an iFixIt kit that came with pointy end tweezers, use these tweezers instead. Remember to twist the grill counter-clockwise to remove it. It does not take a lot of force – but requires good “grip”.