Citation Context

...e distributions – We also consider failure distributions based on failure logs from production clusters. We used logs for the largest clusters among the preprocessed logs in the Failure trace archive =-=[17]-=-, i.e., for clusters at the Los Alamos National Laboratory [26]. In these logs, each failure is tagged by the node —and not just the processor— on which the failure occurred. Among the 26 possible clu...

Citation Context

... distributions – We also consider failure distributions based on failure logs from production clusters. We used logs from the largest clusters among the preprocessed logs in the Failure trace archive =-=[18]-=-, i.e., from clusters at the Los Alamos National Laboratory [24]. In these logs, each failure is tagged by the node —and not just the processor— on which the failure occurred. Among the 26 possible cl...

Citation Context

...rids and clouds reside in two extremes of the reliability and cost spectrum. Grid resources are often regarded as unreliable. Studies [1]–[3] and empirical data collected in the Failure Trace Archive =-=[2]-=- give strong evidence of the low long-term resource availability in traditional and desktop grids, with yearly resource availability averages of 70% or less. The constrained resource availability in g...

Citation Context

...istributed systems. Even for these studies, the findings are based on data corresponding to a single system– until the recent creation of online repositories such as the failure Failure Trace Archive =-=[14]-=- and the Computer Failure Data Repository [5], failure data for distributed systems were largely inaccessible to the researchers in this area. The occurrence of failure bursts often makes the availabi...

Citation Context

...e distributions – We also consider failure distributions based on failure logs from production clusters. We used logs for the largest clusters among the preprocessed logs in the Failure trace archive =-=[30]-=-, i.e., for clusters at the Los Alamos National Laboratory [16]. In these logs, each failure is tagged by the node —and not just the processor— on which the failure occurred. Among the 26 possible clu...

Citation Context

... next failure [12], [19]. Dynamic programming heuristics and simulation results are provided in [12] for single-task applications, using either synthetic traces or failure logs of production clusters =-=[20]-=-. VII. Related work There is a large body of literature on checkpointing strategies for divisible jobs. The corresponding scheduling problem is to partition the job into several chunks and to checkpoi...

Citation Context

...h Markov models using network-wide statistics as in [26, 30]. Unfortunately, some assumptions such as homogeneous behavior, or memoryless exponential distributions are not verified in certain systems =-=[14, 5, 20]-=-. Instead, in this paper, we propose to trigger a repair process by inference of node failure rather than conclusive knowledge. This suggests that one might analyse node failure detection in a standar...