Twitter and Joyent: Update
By David Young 31 January 08 — Twitter has been officially off Joyent since 10PM last night. This may come as a surprise to some after yesterday's posts here and here regarding the two companies working together. Those of us at Joyent appreciate …
Context & Ripple Effects
Twitter's reliability problem has been building in public for months: July 2007 coverage framed the service as having been 'Pwnced' while rival Jaiku waited on its scaling failures, and December brought Dave Winer asking outright why Twitter keeps going down. By late January 2008 the outages were confirmed to be increasingly frequent, including a fall during a Steve Jobs keynote.
What changed this week is structural, not just operational: Joyent's own blog, run by David Young, confirms Twitter has been officially off Joyent's infrastructure since 10PM on January 30 — a reversal of tone after posts the day before describing the two companies working together. The split landed amid a rough stretch: two outages on debate day alone (one hitting during the Republican Primary debate), plus an awkwardly timed glowing review of its hosting provider published on the Twitter blog shortly before the outage. The story travelled fast, picked up same-day by the Twitter Blog ('We had a rough night'), Scripting News ('Why does Twitter go down?') and Dembot.
First-order effects
- Twitter now owns its entire serving stack with no hosting partner to share blame or load — every subsequent outage lands squarely on Twitter's own operations, at exactly the moment the service has become a confirmed real-time channel for political discussion during primary season.
- Joyent loses its highest-profile reference customer overnight, and the whiplash between yesterday's 'working together' posts and today's confirmation puts David Young's account of the relationship under immediate scrutiny.
Second-order effects
- Jaiku, which mid-2007 coverage positioned as readying its own push while Twitter stumbled, gets a fresh opening among users burned by repeated evening outages at exactly the moments they want the service most.
- The public split turns hosting reliability into a marketing battleground: other infrastructure providers now have a cautionary tale to sell against, and Twitter must prove self-managed infrastructure beats what it left behind.
Third-order effects
- If outages keep clustering around live events — keynotes, debates — real-time messaging will be judged on uptime during traffic spikes rather than features, forcing consumer services of this kind to treat operations engineering as a core competency rather than a rented one.
- Twitter's confirmed plan to launch a dedicated Japanese version multiplies the stakes: international expansion assumes the reliability question is solved first, and failure there would set the template for whether such services can scale globally.
The trend: Real-time consumer messaging services are being stress-tested by live-event traffic, shifting the competitive battleground from features to owned infrastructure and uptime.