Update Makes Server Replacement Critical
I have an old Sun server, with an integrated management console that uses SSH for access. That is, until the SSH client removed support for the old encryption used by the server.
I recently retired my Sun T5220 as it developed an error on one of the drives that prevented it from booting. Rather than spend the time to try to fix the old beast, I moved the databases to other servers and unplugged the old box.
My other big acquisition that has aged itself out is the massive Sun X4600-M2 I have. When I got it almost 10 years ago (wow...that long already?) it was already a little old, past its EOL, but it was still ahead of other computers, especially for the couple hundred bucks I spent for it on eBay. I threw a bunch of drives in it, and with its 8 quad-core CPUs and 128GB of RAM, it's been happily running my server ever since. I poked around with Solaris on it for a bit, but ultimately decided to use Linux instead.
Today, I was doing a little software poking around, and when I tried to push something to my git repo, I got the error message that warned my SSH config file contained problematic config:
Bad key types 'ssh-rsa,ssh-dss'
That was added to allow my more modern workstations to access the older server control boards. It was necessary to connect to the older server's Solaris, which never got an updated set of crypto algorithms; this server was not directly connected to the Internet, though. For the big server, which is connected to the Internet, it was only needed to get to the integrated manager (which is not connected to the Internet), as the OS is more modern and does have updated crypto.
I still have a couple SSH tools that still support the older crypto, but if those also get updated, I won't be able to control the server.
I've been looking at some inexpensive replacements. I'm not doing the "big deal" things I was trying to do before, so I don't think I need all the bits and bobs this offers. The only time it approaches full utilization is immediately following boot-up, when it's starting everything and it rebuilds a bunch of caches. I've only purposefully been able to blast its CPUs when trying to use BONIC to play with protein folding, as I did to feel like I was doing something to help the development of COVID vaccines. Even then, it's never come close to filling its RAM. Even with all of the database servers running, and generous memory configuration, there just isn't the kind of traffic necessary to utilize it. I thought to run some generative AI on it, but the CPUs are too old, and the hyperthreading evidently isn't enough.
I've got another internal server with a 16-core (probably 8 core, dual threads each) Ryzen 7 and 64GB of RAM that outperforms the big server, and still has RAM to spare. Its running a couple instances of generative AIs to experiment with. They hog the RAM, and during some queries, the CPU gets a bit busy, but it's always a short lived burst of activity for the seconds it takes to do its thing.
Given the kinds of things it does, and the load it's under, I'm looking at a few different MiniPCs to replace the old server. They tend to be built on laptop-class CPUs, but they almost all beat the old Athlon CPUs in the X4600. There are a few Ryzen 7 boxes that support or come with 64GB of RAM or more for about what I paid for the Sun server. Plus they sip electricity, compared to puling through 4x900W PSUs (usually drawing less than 100W on either UPC where they're attached).
I have the smallest concern about the storage capabilities in the MiniPCs, but the server is using less than 100GB now, leveraging network storage for the few bigger things. Given the relatively lightweight workloads it has, a couple of TB SSDs would exceed what it's using by an order of magnitude.
I've also considered that I could get a few of these and cluster my Docker and AI experiments. The lower-end units would still exceed what I need for the static web, app, and databases the server mostly hosts. Adding a couple higher-end boxes for the AI could allow either multiple LLMs at the same time, or distributing the generation for faster responses. I don't really need anything faster, but it always eats at a person.
Regardless, with the change to the SSH client, not being able to connect to the control module means I can't start the server in the event of a power loss. If my tablets' clients also lose that ability, it's just a big, heavy brick.