Using Fault Injection and Modeling to Evaluate the Performability of Cluster-Based Services

4th USENIX Symposium on Internet Technologies and Systems

Kiran Nagaraja ; Xiaoyan Li ; Ricardo Bianchini ; Richard P. Martin ; Thu D. Nguyen

Others  :  http://www.usenix.org/publications/library/proceedings/usits03/tech/full_papers/nagaraja/nagaraja.pdf
PID  :  8899

来源: CEUR

【 摘 要 】

We propose a two-phase methodology forquantifying the performability (performance and availability)of cluster-based Internet services. In the firstphase, evaluators use a fault-injection infrastructure tomeasure the impact of faults on the server’s performance.In the second phase, evaluators use an analytical modelto combine an expected fault load with measurementsfrom the first phase to assess the server’s performability.Using this model, evaluators can study the server’ssensitivity to different design decisions, fault rates, andenvironmental factors. To demonstrate our methodology,we study the performability of 4 versions of the PRESSWeb server against 5 classes of faults, quantifying theeffects of different design decisions on performance andavailability. Finally, to further show the utility of ourmodel, we also quantify the impact of two hypotheticalchanges, reduced human operator response time and theuse of RAIDs.