> I have an RPi4 as a home server now. It's not a supercomputer...

Out of curiosity, I looked up the performance numbers for the RPi4 and the Cray-1 supercomputer, the world's fastest computer in 1977. The RPi4 gets 748 megaflops on the LINPACK benchmark, compared to 14 megaflops on the Cray. The Cray used high-speed ECL logic allowing it to run at a blistering 80 Mhz clock rate compared to 1.5 GHz on the RPi4. The Cray-1 sold for $7.9 million, compared to $35. It had 8 megabytes of RAM and weighed 5.5 tons compared to 2 GB of RAM and 46 grams. Power consumption was 115 kW vs under 15 watts. The numbers drive home just how much computer technology has improved.

To be sure! Our expectations of what a computer required to do $task looks like has dramatically changed, too. Last week a coworker was surprised that I deployed a tiny service on a t2.nano EC2 instance. "Shouldn't we put that on something bigger?" Well, the thing is entirely blocked on waiting for network IO, so putting it on a server with 32 cores isn't going to make it run any faster. Also, under full load it only used 32MB of RAM, so giving it 8GB isn't going to help it any.

Bringing it back to the RPi: people ask if it's powerful enough to serve simple web apps and email. Oh, it most certainly is! It can handle most of what you'd want a home server to do, short of something like Plex transcoding.

One of my first jobs in 1997 was as a network admin, setting up an email, web server and firewall for our 100 person office. Used a 486 DX2/66 with 16MB (not GB!) RAM running RedHat, obviously very minimal without X or GUIs. The Internet line was only 128kb/s, so the CPU was underutilized most of the time.

Hah, much the same here. We were on some sweet Pentium 133s on Debian, and even our blistering fast 4 bonded T-1s weren't enough to make them break a sweat.

I know MHz isn't everything, but a 1.5GHz quad-core ARM is abundantly sufficient for running Wordpress and Django and a PostgreSQL backend with plenty of room to spare.

Why not Lambda then ?

It's a long-running, stateful process. I love Lambda but this wouldn't have been a good fit for it.

> The numbers drive home just how much computer technology has improved.

And how much bloated software has become.

Software bloat is part of it. The other part is that we just continue to use slow languages. Using python in production for large servers should be illegal due to wasted power.

Why? Most servers are network bound, not CPU bound.

Do you have any sources that modern servers which host dozen and maybe even hundreds of virtual machines are network bound? That doesn't seem logical at all.

What is most amazing: what people achieved with those 14 megaflops.

In your experience, would an external SATA 2.5" SSD in a cheap cradle work faster than the MicroSD?

absolutely.

What do you mean by onboard storage? It uses microSD for storage, right?

Yeah. I use a Raspberry 2 for my mail server, but I use an SSD in a USB enclosure for mail spool. It's also faster, but above all, it's more robust than SD cards.

USB on all Raspberry Pi below 4 is capped at 480 mbit/sec due to it being USB2. Pi 4 has USB3, but it can interfere with 2.4 GHz signals.

USB2 is still faster than all but V90 class MicroSD cards. And V90 cards are only rated for 90MB/s writes, not reads.

Even if you had a UHS-III V90 A2 card, a USB2 SSD is going to give you 10x the IOP/s performance.

Even the USB2 speeds are overkill for most server stuff, where you're more commonly making a bunch of small random accesses than long streaming reads or writes. Not that you can't, but most people are going to be running things like Wordpress or other relatively lightweight services on an RPi.

yes, the microSD card isn't known for being super speedy (but it's not absolutely terrible either).

mSD is slower than an SSD.

Which SD card class?

I bought an all-in-one kit (https://www.amazon.com/gp/product/B07VFCB192) with a Samsung EVO+ 32GB class 10 SD card. To be clear, it's not awful. For most stuff it's perfectly usable. Still, stuff like software upgrades and random server IO is a lot slower than on other media.