Boffins warn LIMPWARE takes the pleasure out of cloud
One flaccid node and the whole cloud goes soft
Computer science boffins from the USA have come up with a lovely term to describe under-performing hardware: “Limpware”.
The term's not just for fun, but actually has some pretty serious implications for cloud computing because the boffins have run tests suggesting just one limp node in a cloudy cluster can lower performance of the entire rig.
Limpware's now been described in three papers. The first, Limplock: Understanding the Impact of Limpware on Scale-Out Cloud Systems (PDF) defines Limpware as “hardware whose performance degrades significantly compared to its specification.”
Plenty of hardware does that, but the authors believe the large loads imposed on clouds make it more likely that Limpware will develop. They also write that they have “collected reports that show how disk and network performance can drop by orders of magnitude” if a single piece of limpware is present in a cloud rig.
The authors term such performance hits “Limplock” and in a second paper, The Case for Limping-Hardware Tolerant Clouds (PDF) argue that “current systems fail to properly manage limping hardware, and thus performance failures often cascade.”
A third paper, Impact of Limpware on HDFS: A Probabilistic Estimation (PDF) offers a detailed analysis of how a single limplocked component, in this case a single NIC card, can greatly degrade the performance of a Hadoop cluster. The paper also shows that Hadoop can't detect the under-performing NIC and therefore doesn't fail over to another. The two other papers offer similar evidence for performance hits and failure to detect limplocked components in HBase, Cassandra, and ZooKeeper.
The three papers collectivity call for more studies of limpware and limplock, so that cloud software can be built to work around the problem.
Those suggestions appear rock solid. ®