I worked at several x86 Linux + SPARC Solaris shops between 1999 and say 2011. Linux was always on the app servers, and Solaris on the DB servers.
The Sun hardware was just better, more robust, and the machines tended to have hot-swappable bits. Better support for fast storage. Hot-plug in Linux was bad and took time to get good. The hardware was cheap, and took time to get good. Ditto driver support. It just got better and better until there was no reason to buy Sun.
And then Oracle bought Sun, and there was now a reason to _avoid_ Sun.
In the PNW, eskimo.com was using Sun machines at least in 1994 when I joined, and I assume for some years earlier. Oz.net was using Irix on SGI machines :)
That stuff was surely popular, too! But we were running LAMP stacks in the late 90s to host customer content on cheap x86 boxes, and it that was an enormously popular hosting solution for many years before 2005.
Sun boxes were very nice machines, but an entry level Sun Fire V480 debuted for $20K, and that would buy a whole tabletop of x86 servers in tower cases.
There was a much greater variety of plausible server options back then, to be sure. I'm mainly arguing against the idea that Linux+x86 was useless until 2005 or so. I had personally worked in 5 different ISP/hosting companies by then which all used that exact combination.
Oh, absolutely, fair point. I used linux exclusively on the desktop from 95-02.
Even commercially; I worked at a decent-sized digital services company in 99-02 that, from the day I started, had 2 ALR 6x6 pentium pro machines as database servers (6 proc, 6 hot swap drive bays). When they crashed, our main issues were with really long-running `fsck` because journaling filesystems were not a thing.
All the app servers were white label intel boxes. We had issues, sure -- the one that comes to mind chiefly is that we were doing IP-based virtual hosting (I don't think name-based virtual hosting was a thing yet), and Linux seemed to get unstable and randomly drop the virtual interfaces once you exceeded maybe a few hundred per NIC, and you'd have to restart the i/f to fix it. I don't think these were behind LBs yet, but I can't really remember.
All that stuff was on RedHat, the first time of 2 or 3 times that Redhat went through the v7 -> v8 -> v9 period :)
Even in much later years (eg, 2008-ish), I remember that too many vendors (HP, Dell, etc) would ship these prosumer grade RAID cards that absolutely fell over (locked up) at sustained high util %. You could (probably correctly) argue that was because we didn't pony up for the true high-end x86 hardware, but the fact that enterprise server companies shipped this stuff at all meant it made the x86 option look less robust compared to the big iron.
