Twitter meltdown raises questions about site stability
Micro-blogging site knocked over by stiff burst of wind
The paralysing effect of an internet attack against Twitter has raised questions about the site's apparent fragility.
Attacks against accounts maintained by pro-Georgian blogger Cyxymu at a number of social networking sites including Facebook, Blogger and LiveJournal as well as Twitter, and apparently aimed at silencing him, brought the micro-blogging site to its knees.
The attack caused intermittent difficulties accessing Facebook (see notice here) and other sites on Thursday, but it was over at Twitter where it really hit home, flooring the micro-blogging service for almost two hours and reducing service levels well into Friday.
Bill Woodcock, research director at Packet Clearing House, advanced the theory on Thursday that the assault wasn't the result of a traditional distributed denial of service, but the effects of users clicking a link contained in spam messages ostensibly promoting Cyxymu's web presence.
The messages were designed to discredit Cyxymu by associating him with a spam run. Other security researchers, such as Patrik Runald at F-Secure (here) and Graham Cluley at Sophos, are sceptical about this Joe Job-style theory for the attack.
The vast majority of recipients wouldn't have bothered clicking on such a link, but it is possible that the spam campaign was either run alongside a denial-of-service attack from a network of compromised PCs or inspired a Russian patriot with access to a botnet to attack Cyxymu's web presence and by extension the social networking sites he uses. The timing of the attack coincides with the first anniversary of the ground war between Russia and Georgia.
However the attack was caused, and whether or not there's any significance in its timing, there's little doubt that it succeeded in throttling Twitter. An analysis by Arbor Networks, experts in DDoS attack mitigation, explains that Twitter-related traffic slowed to a trickle.
We generally don’t see a lot of data (i.e. it takes thousands of tweets to match the bandwidth of a single video), but 55 ISPs in the Internet Observatory were exchanging roughly 200 Mbps with Twitter before the DDoS. Then traffic dropped to a low of 60 Mbps around 10:40am and began climbing after that. As of 1pm EDT, Twitter traffic was still down by 50% at 150 Mbps (normally we see close to 300 Mbps for this time of day).
Twitter’s two NTT hosted address blocks were moved in response to the attack, Arbor adds. Twitter's reliance on just one service provider, and apparent lack of back up and redundancy, much less a comprehensive disaster recovery plan, goes a long way towards explaining why it was hit so badly.
Twitter's website was back up and running, albeit with minor latency issues, by Friday. The latest status update from Twitter states that "site latency has continued to improve, however some web requests continue to fail". ®