Search

Exascale studies project reliability challenges for future HPC systems. We present the Global View Resilience (GVR) system, a library for portable resilience. GVR begins with a subset of the Global Arrays interface, and adds new capabilities to create versions, name versions, and compute on version data. Applications can focus versioning where and when it is most productive, and customize for each application structure independently. This control is portable, and its embedding in application source makes it natural to express and easy to maintain. The ability to name multiple versions and “partially materialize” them efficiently makes ambitious forward-recovery based on “data slices” across versions or data structures both easy to express and efficient. Using several large applications (OpenMC, preconditioned conjugate gradient (PCG) solver, ddcMD, and Chombo), we evaluate the programming effort to add resilience. The required changes are small (< 2% lines of code (LOC)), localized and machine-independent, and perhaps most important, require no software architecture changes. We also measure the overhead of adding GVR versioning and show that overheads < 2% are generally achieved. This overhead suggests that GVR can be implemented in large-scale codes and support portable error recovery with modest investment and runtime impact. Our results are drawn from both IBM BG/Q and Cray XC30 experiments, demonstrating portability. We also present two case studies of flexible error recovery, illustrating how GVR can be used for multi-version rollback recovery, and several different forward-recovery schemes. GVR’s multi-version enables applications to survive latent errors (silent data corruption) with significant detection latency, and forward recovery can make that recovery extremely efficient. Our results suggest that GVR is scalable, portable, and efficient. GVR interfaces are flexible, supporting a variety of recovery schemes, and altogether GVR embodies a gentle-slope path to tolerate growing error rates in future extreme-scale systems.

In a series of 24 cases of acute dissecting aneurysm of the aorta (not including Marfan's disease) the diagnosis was usually suspected on the basis of the clinical picture and plain chest roentgenograms. The most consistent clinical sign was severe pain. Absent pulses and a neurological deficit were each noted in only five patients. In many cases there was no correlation between the clinical picture and the type or the extent of the dissection. Widening of the aortic arch and obliteration of the aortic knob with displacement of the trachea to the right are the most common signs in plain chest roentgenograms. A barium swallow examination in these cases reveals an elongated compression and displacement of the esophagus by the aortic arch. Calcification in the area of the aortic arch is the exception rather than the rule in dissecting aneurysms. Angiography is essential for the definitive diagnosis of dissecting aneurysms. The diagnosis is based on the demonstration of two channels, either by the presence of a linear radiolucency separating the two lumens, or by differences in flow that present as delayed opacification or delayed washout. If only the true lumen is opacified, widening of the outer extraluminal border of the aorta or narrowing of the lumen indicates the presence of a dissection. Abnormal catheter recoil and position were helpful in only two cases, and are not informative when the false lumen is catheterized. Failure to visualize main aortic branches was not always due to involvement by the dissection. It can also be caused by reduced flow due to severe proximal compression of the main lumen. The exact location of the intimal tears is usually not demonstrated unless additional injections are made in the area assumed to contain the tear. If only the false lumen is opacified in the ascending aorta, this can be recognized by the demonstration of a blind end, by failure to visualize the sinuses of Valsalva, from flattening of the medial border of the opacified channel, and from delayed washout in the blind end.