C. Basile, Z. Kalbarczyk, K. Whisnant, and R. K. Iyer. Active replication of multithreaded applications. Technical Report CRHC-02-01, University of Illinois, 2002.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....authors in [25, 24, 2] address fault tolerance in the context of distributed shared memory (DSM) machines but their approach requires somewhat extensive hardware support. Software based approaches to building fault tolerant systems of commodity, off the self components include active replication [3], development of fault tolerant management software libraries [14, 29] and application transparent techniques, such as our own. Due to space limitations, we shall next focus on the latter category. Sultan et al. 27] follow a log based approach to tolerate single node failures in a home based, ....

C. Basile, Z. Kalbarczyk, K. Whisnant, and R. Iyer. Active replication of multithreaded applications. Technical Report CRHC-02-01, University of Illinois at Urbana-Champaign, 2002.

....and an External Correctness property, which defines the behavior of the algorithm with respect to other replicas. These two properties are formally defined below. This is relatively simple for blocking I O operations. In our replication framework, it is done by using the Virtual Socket Layer [4]. The temporal logic symbol S denotes since p S q indicates that q was true in the past and p has been true from that moment until now. Given a predicate p, p holds infinitely often indicates that from now on p holds an infinite number of times (this is denoted with p by using linear ....

....to a voter fanout process and the other connecting the voter fanout process to three replicas. Replicas and voter execute on Pentium III 500 MHz based machines running Linux 2.4, and Ensemble 1.38 [13] is used for group communication. The replication framework employed is the Virtual Socket Layer [4], which provides transparent active replication to socket based applications. Table 1. Services Description. Service Type Sequence of Activities Scenario A lock m 0 lock m 1 unlock m 0 unlock m 1 I O lock m 2 unlock m 2 I O Combination of accesses to a two and one level data ....

C. Basile, Z. Kalbarczyk, K. Whisnant, and R. Iyer. Active replication of multithreaded applications. Technical Report CRHC-02-01, University of Illinois at Urbana-Champaign, 2002. http://www.uiuc.edu/cbasile/lsa.

....m (proj q[m] that stores the subsequence of H corresponding to mutex acquisitions on mutex m that have yet to be enforced. The follower invokes the function on recv mt upon receiving a mutex table from the leader to append the In our implementation, this is done by the virtual socket layer [4]. We use the linear temporal logic symbol to denote eventually. In absence of failures, in lsa lock the lines 21 23 are not executed, and the condition at line 25 is always true; in on recv mt the condition at line 3 is always false; on leader failed and reconf are not invoked. ....

.... mt lines 10 12) or (2) proj q[m] contains an entry (m; t ) with k = index(m; t ) immediately preceeding (m; t) and thread t releases m, its k mutex acquired, through lsa unlock (lsa unlock lines 6 8) Due to the space limitation, formal proofs of LSA correctness are relegated to [4]. 5. Failure Behavior with Error free Leader toFollowers Communication The LSA algorithm introduces asymmetry in replicas (leader and followers) and requires direct communication from leader to followers. This brings about failure modes not present in traditional replication schemes (e.g. ....

[Article contains additional citation context not shown here]

C. Basile, Z. Kalbarczyk, K. Whisnant, and R. Iyer. Active replication of multithreaded applications. Technical Report CRHC-02-01, University of Illinois at Urbana-Champaign, 2002. http://www.uiuc.edu/cbasile/lsa.

....mutex acquisitions present in both replica histories, if they are causally related in one replica history, they are causally related, in the same way, in the other replica history; and (liveness) if a mutex acquisition is in one replica history, it will eventually be in the other replica history. [3] formally shows that the LSA algorithm maintains correctness between any pair of nonfaulty replicas (e.g. leader follower, follower follower) 2.2. Failure Free Behavior The leader records its mutex acquisitions in a buffer (mutex table) which is reliably multicast to the followers. ....

....(which is the only communication This is analogous to the notion of causal precedence between two events in a distributed system. An analysis under Byzantine behavior of the reliable multicast protocol (e.g. a leader sending different mutex tables to different followers) can be found in [3]. from the leader to followers) Since the reliable multicast does not fail, all nonfaulty followers receive the same sequence of leader s messages, even if the leader sends corrupted data. This guarantees correctness for each pair of nonfaulty followers. Hence, followers grant the same causally ....

[Article contains additional citation context not shown here]

C. Basile et al. Active replication of multithreaded applications. Technical Report CRHC-02-01, University of Illinois at Urbana-Champaign, 2002.

No context found.

C. Basile, Z. Kalbarczyk, K. Whisnant, and R. K. Iyer. Active replication of multithreaded applications. Technical Report CRHC-02-01, University of Illinois, 2002.

No context found.

C. Basile. Active replication of multithreaded applications. Technical Report CRHC-02-01, University of Illinois at Urbana-Champaign, 2002.

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