| This part caught my interest: > each server (so called “shards”) ran on multiple Solaris machines (the map was split by regions). Found this 13 year old Quora commment (from https://www.quora.com/What-was-the-technology-stack-driving-...). Ruben Cortez Those original servers were Sun Ultra II's running Solaris. If I remember correctly, it was a dual proc box (rare in those days), 300mhz, with 256MB of ram (which we subsequently upgraded to 512MB for a boat load of cash). The server weighed a ton, and certainly wasn't rack mountable. and they cost about $30k each, making one UO shard cost about $150k (not including external storage), which is a ridiculous amount of money these days for a single shard. We eventually built MMO shards for a lot less (including MCO, TSO, ENB, and of course SWTOR) cost per shard. Back ups were kept on internal storage, for which we sweated buckets over everytime we lost a drive. we eventually centralized backups on an Hitachi Storage array that was a monster and weighed easily 100lbs and provided a whopping ~10GB of raw space. I think where we revolutionized administration of a distributed MMO was on the network side, specifically via the VPN, which was very new at that time. Short of ordering PTP circuits all over the country and world, we used a software VPN to create those tunnels over our public internet connection at the time (a SINGLE DS3 -- it wasn't until late 1998 that we had a second circuit!) to allow for login handoff, administration, backups, publishes, etc. our hub and spoke VPN design and subsequent fully-meshed design were what made distributing shards economically feasible. I think it was around mid 1999 when we converted to Linux, but not before using Solaris x86 first for a time. We bought Dell Towers to act as servers (Dell didn't ship a true rack-mountable server until the following year or so I think) and they were slightly better on speed than the Suns and at a much lower cost. But i think we needed more of them per shard, especially as we released expansion packs (which in those days, meant adding a new server to handle the added land mass). We were likely closer to 10 servers/shard around 2000/1. Credit to Mark Rizzo for the architecture and buildout of the UO's backend. It was way ahead of it's time, and the innovations we mustered back then is so taken for granted these days (but isn't that true for everything?!). That timeline for Solaris (Sparc) -> Solaris (x86) -> Linux (x86) feels very familiar from my small company developer job at the time. |