K. Nagaraja, X. Li, B. Zhang, R. Bianchini, R.P. Martin, and T.D. Nguyen, "Using Fault Injection to Evaluate the Performability of Cluster-Based Services," Proc. Fourth USENIX Symp. Internet Technologies and Systems (USITS 2003.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... in this paper we examine the impact of communication hardware, resource exhaustion, and application faults on the availability, performance, and performability of PRESS, a cluster based locality conscious Web server [9] The study relies on our own fault injection and analysis methodology [26] to examine the differences between TCP and VIA, according to their implementations over the Giganet cLAN network. Our study illustrates several performance vs. robustness tradeoffs. We show that TCP, in spite of its sophisticated time out and retry, does not provide greater availability than a ....

....VIA substrates, we suggest directions for the design of future communication layers. For example, we believe that a high performance, robust communication layer should be message based, single copy, and pre allocate all resources. 2. Methodology We will use the two phase methodology proposed in [26] to evaluate the performability of PRESS in the presence of faults. In the rst phase, the evaluator denes the set of all possible faults, then injects them (and the subsequent recovery) one at a time into a running system. During the fault and recovery periods, the evaluator must quantify ....

[Article contains additional citation context not shown here]

K. Nagaraja, X. Li, B. Zhang, R. Bianchini, R. P. Martin, and T. D. Nguyen. Using Fault Injection to Evaluate the Performability of Cluster-Based Services. Technical Report DCS-TR-491, Department of Computer Science, Rutgers University, May (Revised August) 2002.

.... in this paper we examine the impact of communication hardware, resource exhaustion, and application faults on the availability, performance, and performability of PRESS, a cluster based locality conscious Web server [9] The study relies on our own fault injection and analysis methodology [26] to examine the differences between TCP and VIA, according to their implementations over the Giganet cLAN network. Our study illustrates several performance vs. robustness tradeoffs. We show that TCP, in spite of its sophisticated time out and retry, does not provide greater availability than a ....

....VIA substrates, we suggest directions for the design of future communication layers. For example, we believe that a high performance, robust communication layer should be message based, single copy, and pre allocate all resources. 2. Methodology We will use the two phase methodology proposed in [26] to evaluate the performability of PRESS in the presence of faults. In the first phase, the evaluator defines the set of all possible faults, then injects them (and the subsequent recovery) one at a time into a running system. During the fault and recovery periods, the evaluator must quantify ....

[Article contains additional citation context not shown here]

K. Nagaraja, X. Li, B. Zhang, R. Bianchini, R. P. Martin, and T. D. Nguyen. Using Fault Injection to Evaluate the Performability of Cluster-Based Services. Technical Report DCS-TR-491, Department of Computer Science, Rutgers University, May (Revised August) 2002.

....modeling to quantify FME s potential for increasing PRESS s availability. At this point, we do not yet have an implementation of FME. Rather, our quantification is computed on a hypothetical FME infrastructure. Our work is motivated by a recent effort to study the behavior of PRESS under faults [11]. We found that PRESS performed poorly even when subjected to only a modest range of common faults, including disk hangs, link and switch failures. We soon concluded that the designer s implicit fault model did not match the faults we were injecting. While it was certainly possible to improve ....

....complex fault model is also extremely difficult, if not impossible. Thus, after many more fault injection experiments, we decided that reducing the possible state space using FME was a necessary and viable approach for dealing with this complexity. We use a methodology that we recently introduced [11] to evaluate the performability of several versions of PRESS when running with and without FME. This methodology combines measured responses to fault injection and analytic modeling. The results of the study show that a PRESS server following an FME approach has the potential to decrease the ....

[Article contains additional citation context not shown here]

Kiran Nagaraja, Xiaoyan Li, Bin Zhang, Ricardo Bianchini, Richard P. Martin, and Thu D. Nguyen, "Using Fault Injection to Evaluate the Performability of Cluster-Based Services," Tech. Rep. DCS-TR-491, Department of Computer Science, Rutgers University, May 2002.

.... in this paper we examine the impact of communication hardware, resource exhaustion, and application faults on the availability, performance, and performability of PRESS, a cluster based locality conscious Web server [10] The study relies on our own fault injection and analysis methodology [34] to examine the differences between TCP and VIA, according to their implementations over the Giganet cLAN network. Our study illustrates several performance vs. robustness tradeoffs. We show that TCP, in spite of its sophisticated time out and retry, does not provide greater availability than a ....

....VIA substrates, we suggest directions for the design of future communication layers. For example, we believe that a high performance, robust communication layer should be message based, single copy, and pre allocate all resources. 2. Methodology We will use the two phase methodology proposed in [34] to evaluate the performability of PRESS in the presence of faults. In the first phase, the evaluator defines the set of all possible faults, then injects them (and the subsequent recovery) one at a time into a running system. During the fault and recovery periods, the evaluator must quantify ....

[Article contains additional citation context not shown here]

K. Nagaraja, X. Li, B. Zhang, R. Bianchini, R. P. Martin, and T. D. Nguyen. Using Fault Injection to Evaluate the Performability of Cluster-Based Services. Technical Report DCS-TR-491, Department of Computer Science, Rutgers University, In preparation.

No context found.

K. Nagaraja, X. Li, B. Zhang, R. Bianchini, R.P. Martin, and T.D. Nguyen, "Using Fault Injection to Evaluate the Performability of Cluster-Based Services," Proc. Fourth USENIX Symp. Internet Technologies and Systems (USITS 2003.

No context found.

K. Nagaraja, X. Li, B. Zhang, R. Bianchini, R. Martin, and T. Nguyen. Using Fault Injection to Evaluate the Performability of Cluster-Based Services. In 4th USENIX Symposium on Internet Technologies and Systems (USITS), Seattle, WA, March 2003.

No context found.

K. Nagaraja, X. Li, B. Zhang, R. Bianchini, R. Martin, and T. Nguyen. Using Fault Injection to Evaluate the Performability of Cluster-Based Services. In 4th USENIX Symposium on Internet Technologies and Systems (USITS), Seattle, WA, March 2003.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC