Growing user demands have placed increasing pressure on the company’s hardware. In 2004, EVE Online switched to solid-state disks in order to avoid the delays introduced by moving mechanical parts in traditional hard drives. In 2006, the company upgraded to 64-bit computing to achieve higher-density storage. CCP Games is now moving its clusters over to IBM blade servers, a system of servers on cards that allows for greater storage density, power efficiency, and communication between servers. Petursson says that he hopes the company’s new supercomputing clusters will add enough capacity for 70,000 concurrent users.
George Dolbier, the global technical lead for games and interactive entertainment at IBM, says that to scale up to support millions of concurrent users, MMOs will need to make use of the technologies behind Visa’s database, or those that support the NASDAQ stock market. NASDAQ, he says, can actually be thought of as a very large MMO, supporting very large numbers of “players” performing billions of transactions daily in a graphically intense environment, all within a single shard.
Large-scale MMOs, Dolbier says, share many computing problems with the financial industry. MMOs, like financial institutions, require high-speed communication between servers, since delays frustrate customers and provide opportunities for cheating and fraud. MMOs also require the maintenance of very large databases. EVE Online’s servers, for example, which now support only 200,000 players, currently process more than 150 million operations per day.
Dolbier says that it’s not a one-way street: other industries could learn from the technology that’s being developed for MMOs. For example, he says, a common problem for game companies is how to recognize and manage “hot spots”: small areas that suddenly attract large numbers of players, such as a battleground. To keep the game running smoothly, the servers need to detect movement toward the hot spot and react in real time, rezoning the area of activity so that more servers are responsible for supporting it. Technologies that solve this problem effectively, Dolbier says, will have applications in any industry that requires spotting and reacting to trends, or “anything where behavior is dynamic and you need to move resources around rapidly.”