Exploring Graph Analytics for Cloud Troubleshooting
Chengwei Wang, Karsten Schwan, Brian Laub, Mukil Kesavan, Ada Gavrilovska · International Conference on Autonomic Computing · 2014
We propose VFocus, a platform which uses streaming graph analytics to narrow down the search space for troubleshooting and management in large scale data centers. This paper describes useful guidance operations which are realized with graph analytics and validated with representative use cases. The first case is based on real data center traces to measure the performance of troubleshooting operations supported by VFocus. In the second use case, the utility of VFocus is demonstrated by detecting data hotspots in a big data stream processing application. Experimental results show that VFocus guidance operations can troubleshoot Virtual Machine (VM) migration failures with accuracy of 83% and with delays of only hundreds of milliseconds when tracking migrations on 256 servers hosing 1024 VMs. Such successes are achieved with negligible runtime overheads and low perturbation for applications, in comparison to brute-force approaches.