Cooperative empirical failure avoidance for multithreaded programs

Abstract

Concurrency errors in multithreaded programs are difficult to find and fix. We propose Aviso, a system for avoiding scheduledependent failures. Aviso monitors events during a program's execution and, when a failure occurs, records a history of events from the failing execution. It uses this history to generate schedule constraints that perturb the order of events in the execution and thereby avoids schedules that lead to failures in future program executions. Aviso leverages scenarios where many instances of the same software run, using a statistical model of program behavior and experimentation to determine which constraints most effectively avoid failures. After implementing Aviso, we showed that it decreased failure rates for a variety of important desktop, server, and cloud applications by orders of magnitude, with an average overhead of less than 20% and, in some cases, as low as 5%.

Concurrency debugging with differential schedule projections

“…Request permissions from Permissions@acm.org. ically, Symbiosis produces a DSP from a single failing schedule, enabling its use for failures observed rarely (i.e., in production) and does not require repeated program executions like prior work [31]. Our evaluation in Section 5 shows that DSPs have, on average, 81% fewer events than full schedules and shows qualitatively, with case studies, that DSPs help understand failures and fix bugs.…”

“…in the chain to prevent that failure. This phenomenon is called the Avoidance-Testing Duality, and is detailed in previous work [31]. We now use some case studies to illustrate how differential schedule projections focus on relevant operations and help understand each failure.…”

“…The DSP shows how the two schedules differ in the order of their events and in their data-flow behavior. Additionally, as the reordered pair from the alternate non-failing schedule eliminates an event order necessary for the failure to occur, it can be leveraged by a dynamic failure avoidance system [31] to prevent future failures.…”

“…Unlike these techniques, Symbiosis is not limited to searching for known patterns. Sub-schedule search, in turn, is general and not limited to specific patterns [31,43,52]. Unfortunately, the space of all of an execution's possible sub-schedules can be large.…”

On-the-Fly Repairing of Atomicity Violations in ARINC 653 Software

“…The concurrency error is one of the well-known errors of software [2][3][4][5][6][7][8] and it is famous for its irreproducibility, which makes it very difficult to debug the error [4][5][6][7][8]. In a real-world application, about 70% of all concurrency errors are caused by atomicity violations [3]; it is a concurrent execution of a specific code region, i.e., critical section, that unexpectedly violates that region's atomicity.…”