Engineers at Alibaba Cloud have revealed details of a new recovery service that can “instantly” reroute network traffic during outages.

Dubbed ZooRoute, the system was detailed in a newly published paper, first spotted by The Register, set to be presented at the upcoming SIGCOMM conference.

The engineers behind it contend that traditional methods to alleviate network equipment failures in cloud data centers take too long to reroute traffic around the problem.

Instead, ZooRoute constantly tests different source ports to see which ones lead to working network paths. Once a failure is detected on a path, it changes the source port in outgoing packets to use a different, working path. A post on one of the cloud vendor’s forums from late April suggests ZooRoute can perform disaster recovery “within seconds to ensure connection high availability.”

In their latest paper, Alibaba Cloud engineers revealed ZooRoute was deployed across their entire cloud infrastructure for around 18 months, leading to an alleged reduction in network outage times of more than 92%. Of this, around 98% of the outages were reportedly undetectable by its customers, with cases where the tool failed to kick in largely attributed to “severe network capacity loss and the unstable status of faulty physical devices.”

Instead of simply rerouting traffic during outages, which takes considerable time, Alibaba Cloud engineers designed ZooRoute as a means to actively map out which source ports lead to functioning network paths to steer traffic around problems with much faster response times, selectively probing network paths under normal conditions before launching into more thorough analysis when network failures are detected.

According to the paper detailing the recovery solution, ZooRoute is both tenant-transparent and requires “no modifications to physical devices” – since it manipulates packet headers at the software layer through deployment on host machines rather than network switches.

ZooRoute's success stems from leveraging ECMP (Equal Cost Multi-Path) routing – which spreads traffic across multiple paths by looking at packet headers. The recovery tool takes advantage of ECMP hashing to determine which source ports will route packets to working parts of the network.

Insights into ZooRoute come after Alibaba Cloud researchers recently showcased another tool aimed at improving network management: Eigen+, a smart memory management system that boosts memory utilization in cloud database clusters. Eigen+ uses machine learning to identify “transient” database instances experiencing unpredictable memory spikes to exclude them from over-subscription.

The cloud provider was also one of two Chinese vendors that made it into the “leader” portion of Gartner’s latest Magic Quadrant ranking for container management, receiving accolades for its hybrid cloud offerings and broad functionality coverage.