I used a USB drive to install a NAS system drive, and my 16GB drive only lasted four months

The first night I installed Feiniu on the N100 mini PC, I didn't want to spend extra money on a system drive, so I dug out a spare 16GB Kingston USB drive from a drawer and plugged it into the USB port at the back of the case as a system drive. Everything went smoothly during installation. Feiniu's system drive has a low threshold—8GB is enough, and 16GB is still too much for me. The first few months were peaceful, until one Thursday morning, my wife said the TV couldn't play cartoons on the NAS. I checked my phone and saw the device had been offline all night.
To get straight to the point: it can install and run, but USB drives are just consumables
To put the conclusion upfront: systems like Feiniu and Black Synology can boot, configure, and work properly on a USB drive. After running for a month, you might even feel like you've gained a bargain—saving a drive or dozens of yuan per drive that gathers dust. But the system drive is a job that gets chopped up 24/7, and the flash memory chips and main controller on a regular USB drive are consumables designed to "copy files occasionally." Doing the work of a long-term laborer with consumables, death is just a matter of time. Mine lasts four months, and some lucky enough to last a year. Essentially, there's no difference.
Later, upon review, I realized the problem wasn't the capacity at all. After installing the 16GB, the system is mostly left, and I've never felt a shortage of space. The key issue is the writing method, not the amount written.
What is the system disk secretly writing on its daily routine
?Writing large files on a data drive can quickly take several hundred MB at a time, and even mechanical drives can handle it. The system drive is the opposite, writing the 'small, fragmented, frequent' type
| Approximate | scale of | write sources |
|---|---|---|
| System logs (like /var/log) | are added daily withseveral MB to tens of MB | of additional writes, nonstop all day |
| Docker | container layer checks container quantity, From tens of MB/day | , allparts of the directory that weren't mounted were sent to the system disk |
| The database is small and | the transaction volume is small, but fsync writes are extremely frequent | at 4KB randomly, tens of thousands of times a day |
| Indexes and inspection records | are written in a centralized batch of albumindexes and SMART inspection | scheduled tasks |
These features alone are insignificant; combined, they only generate a few hundred MB of bandwidth per day. My machine ran qBittorrent and albums, and there were no restrictions on container logs (I later calculated this in the article on cleaning storage pools >
How My USB Drive Failed: A Record of That Morning's Investigation
Found out I was missing this morning, I pinged first, but couldn't get through. Connect the monitor to the small host and restart. It gets stuck mid-startup, and a line of 'sda: Write Protect is on' scrolls across the screen—the system drive locks itself to read-only. Restart three times, and all three times get stuck in the same spot.
Unplugging the USB drive and plugging it into the computer, it can recognize the drive and read files, but copying anything inside will report write protection errors and the formatting will fail. When I checked the controller with ChipGenius, the drive was still alive. The controller detected too many bad blocks and proactively switched to protection mode—this was the USB drive's last words: better to lock than write to prevent faster data corruption. I copied the system files inside. I tried using mass-produced tools to flash the main controller, but after flashing, the capacity shrank. Even after reconnecting to the NAS, it wouldn't last a week. There's no need to save a 20-yuan drive.
During those days, the NAS was down, but fortunately, photos and movies were on two data drives. After reinstalling the system, everything came back. The system drive crashed, but the data drive remained intact. This architecture saved me—later, that replacement 64GB SSD also failed once, but reinstalling it still worked fine. That's another story. I wrote an article specifically about If the system drive fails, will data be lost?
Why USB drives can't hold up: three issues: chips, main controller, and balanced wear
With the same flash memory, why can SSDs last ten years, while USB drives die in just four months? Let's break it down and calculate three accounts.
The first is the granules. TLC chips used in reputable SSDs claim a lifespan of 1,000 to 3,000 times; To cut costs, USB drives use more scraps and white chip chips, so their lifespan is more than halved—after a few hundred uses, they're gone.
The second stroke is wear equalization. The SSD controller has a complete FTL (Flash Conversion Layer) that spreads writes across the entire disk and rotates every block, with hotspots automatically spread. Ordinary USB drives either lack wear and tear balance or are extremely minimalist, with system logs and databases repeatedly writing to the same small area—a full drive written twice might be hundreds of times on hotspot blocks. 4KB small random write stacked with fsync is especially ruthless for write amplification, which is torture for a USB drive controller without cache.
The third is job matching. USB drives are designed to be unplugged immediately after copying files, leaving the files idle; The system drive feeds small writes 24/7 × without interruption. Running a marathon with a sprinter—no matter how well the pace is right, they'll still fall.
By the way, I'll teach you how to check the write pressure on your own system drive—don't be like my hindsight commentary. If the SSD is plugged in, go into SSH and type smartctl -a /dev/sda, find the "Total Host Write Volume" line, and divide by the number of days powered on to get the average daily write speed; Then use du -sh /var/log to see how thick the log directory is. If you notice logs taking several gigabytes at a time, or a container is overclocking on the system disk, you can move the directory and add logs to the limit before it is written to your drive.
| is resistant to low write | costs | . When the system drive is low, the conclusion | |
|---|---|---|---|
| that ordinary USB drives | are poor, with no wear and balance | .They can run for 0 yuan (dusty ones), | but can suddenly crash at any time |
| SD card + card reader | Even worse, there's a risk of contact | ,around 20 yuan | , which is even less convenient than a USB drive |
| Small capacity SSDs | are better, main controllers are fully armed | , second-hand 64GB is about 20 yuan | . My final answer |
| mechanical drives are divided into separate zones | , and there's no PE concept | 0 yuan (existing listings) | If you're afraid of noise, don't consider |
more thing: SD cards share the same source as USB drives and are even thinner. Plugging them into a USB card reader for long periods of power generates both heat and poor contact, which is a double debuff. Raspberry Pi fans are used to keeping two SD cards ready to swap them out anytime, but that's for light load scenarios. NAS system setup is a task beyond imagination.
After switching to 64G solid-state states, the world quieted down
I found a second-hand 64GB Kingston SSD for about twenty yuan, specifically as a system drive. After the change, I patched up the previous pitfalls one by one: Docker added size limits to the daemon.json, moved the database and download directories to data disks, and during the hibernation check, I also checked who was still copying secretly. More than a year later, the total write speed of this SSD has long surpassed that USB drive from back then—very stable—the main controller spreads write data evenly, while SmartCTL doesn't budge at all during health checks.
By the way, here's a budget alternative: partitioning the data drive to install the system is also an option. Feiniu can choose this option when building a PC. The downside is that the system data is on the same drive, so if the drive fails, it will be ruined together. Fortunately, mechanical drives have much better write endurance and lifespan than USB drives, making them more reliable than sacrificing a single USB drive. That's exactly what I did with my old laptop remodel. The retired 64GB SSD was directly given to the system installed Feiniu.
Theright role of USB drives in NAS: These three functions are most suitable
After all this, I'm not telling everyone to throw away USB drives. In the NAS community, there are three legitimate positions, all light-loaded, "write once, read for a lifetime."
| How to use | theposition | and why is it suitable |
|---|---|---|
| The RR image of the Black Synology boot disk | is written to the USB drive using Rufus, and | the boot boot guideis read-only, not write-only. Each write lasts five years |
| reset disk | Only insert the USB drive when you need to place the marker file in the FNIOS_RESET | directoryto save your life. Usually, you can lay down |
| and write to the boot disk | to installthe system | and imageas soon as it's done |
The Black Synology setup takes USB drives to the extreme—install USB drives and install the system hard drive. If you want to know, check my previous Black Synology installation guide. Password reset for Feiniu is also a USB drive's job, specifically mentioned in the article about accidental deletion rescue.
If you're going to build up your computer, follow this order: first decide where to place the system (excluding USB drives, choose between SSD or partition), then plan how to arrange the data drives, < a href="https://www.shunot.com/en/nas/949.html"> hard drive selection. After installation, remember to point Docker's logs and database directories to the data drive, so the system drive doesn't take the hit. If you choose the right system setup, you won't think about it for the next three years—that's proof it's doing well.
