On Thu, Apr 30, 2009 at 8:48 PM, Omar Sayeed Khan <omar.sayeed@gmail.com> wrote:
Not only that, I think using RDBM will have a drastical impact on the
Er.. so what is a drastic impact I wonder? If it takes 300 milliseconds longer to connect to IRC, would that be bad? My thinking is most legitimately users generally stay connected to IRC, and rarely reconnect randomly (except when they're done for a while). Usually the Reverse DNS query takes pretty long, >30 seconds, and the answer to an asynchronous DB query would almost certainly arrive back before long the DNS timeout or normal answer... In fact, that would allow the queries to be queued.. If load was too high to immediately answer all queries; the ones that already got a DNS answer could be prioritized, so there would be less noticeable impact on connection time. The users whose queries were delayed are the same as the ones who already delayed connecting (due to their ISP's slow reverse DNS servers) The DNS query _is_ a database query, and it generally has to be made to a server that isn't even local. The primary server resource utilized for a database like this is disk space and disk IOPs to read any un-cached data that is required. I expect the DB would have at most a few hundred thousand entries.. even the most inefficient query strategy should be able to answer thousands of questions in less than a second, on 5-year-old server harware... It's not like IRC servers see millions of unique ips in a limited time period or have 1000+ connections per second per server. -- -J