Limplock
Limplock is sustained slow progress caused by degraded hardware when a system cannot shift affected work to healthy resources.
Scope: Operation, node and cluster limplock distinguish affected requests from wider resource exhaustion. Slow work can occupy shared workers or build queues that obstruct otherwise healthy work. [1]
Detection: A timeout that measures silence can miss continuous but inadequate progress. Retries may revisit the same slow resource, or leave earlier work occupying the capacity needed for recovery. [1]
Isolation: Separating operation classes and destinations can contain propagation. A successful crash-failover test alone does not exercise slow-progress recovery. [2]
Relations
- related
- Fail-slow, Gray failure, I/O timeout, Backpressure
Limitations
- The 2013 experiments concern historical implementations and injected slowdowns, not the incidence of this behavior in current deployments. [1]
Citations
- Do et al. Limplock: Understanding the Impact of Limpware on Scale-Out Cloud Systems. SoCC 2013, sections 3-6.
- Do and Gunawi. The Case for Limping-Hardware Tolerant Clouds. HotCloud 2013, sections 3-5.
Further reading
Sources & caveats
- Full SoCC 2013 paper read, DOI 10.1145/2523616.2523627. Tested Hadoop/HDFS 1.0.4, ZooKeeper 3.4.5, Cassandra 1.2.3 and HBase 0.94.2; 56 experiments across 22 protocols, repeated 3-5 times. Not independently reproduced.
- HotCloud 2013 is the earlier position and exploratory study, not independent corroboration of the later SoCC experiments. Source PDFs retained privately with SHA-256 hashes.