Dynamic Software Updates for Accelerating Scientific Discovery

Abstract

Abstract. Distributed parallel applications often run for hours or even days before arriving to a result. In the case of such long-running programs, the initial requirements could change after the program has started executing. To shorten the time it takes to arrive to a result when running a distributed computationally-intensive application, this paper proposes leveraging the power and flexibility of dynamic software updates. In particular, to enable flexible dynamic software updates, we introduce a novel binary rewriting approach that is more efficient than the existing techniques. While ensuring greater flexibility in enhancing a running program for new requirements, our binary rewriting technique incurs only negligible performance overhead. We validate our approach via a case study of dynamically changing a parallel scientific simulation.

Dynamic software updating: a systematic mapping study

“…Also, number of authors have highlighted additional DSU approaches including dynamic replacement [22, 54, 66, 75, 86, 92, 93 ], dynamic instrumentation [52, 59, 60, 74, 84, 94 ], dynamic linking [37, 39, 53, 75, 83, 85 ], lust‐in‐time compilation [30, 44, 55, 58, 76, 95 ], wrapper [4, 40, 42, 44, 52 ], dynamic aspect‐oriented [35, 41, 50, 96, 97 ], extension of features [98, 99 ], dynamic adaptation [43, 59 ], rolling update [84 ], object transformer [49 ], data‐driven programming [92 ], distributed DSU [38 ], dynamic core library update [56 ], using run time to support DSU [92 ], procedure‐oriented [30 ], micro‐language‐based [100 ], and dynamic rebinding [5 ].…”

“…In addition, number of studies have mentioned additional tools including Jrebel [42, 46, 48, 53, 91 ], LUCOS [22, 51, 76, 79, 94 ], Javelus [35, 40, 42, 44 ], JavAdaptor [16, 38, 46, 52 ], DVM [4, 42, 44, 52 ], HotSwap [16, 42, 50, 57 ], DynaMOS [51, 76, 78, 105 ], Dlpop [37, 39, 106 ], Mx [83, 88, 90 ], EcoDSU [106, 108 ], PROTEOS [51, 95 ], Ekiden [6, 105 ], lusagent [59, 84 ], DYMOS [22, 47 ], EmbedDSU [106, 112 ], UpgradeJ [4, 47 ], Prose [41, 57 ], REPLUS [76, 113 ], FASA [65, 106 ], and Adapt.net [59, 68 ].…”

“…In this examination, we have identified an existing code segment, data segment, stack segment, heap segment, and multithread dynamic updating techniques as follows: A. Code segmentA number of investigations have presented five types of code segment techniques including indirection [4, 46, 47, 53, 55, 74, 77, 94, 99, 106, 119 ], injection [46, 74, 79, 91, 103, 120 ], rewriting [47, 50, 53, 74, 106 ], bytecode manipulation [41, 43, 55 ], and trampolines [4, 106 ]. B. Data segmentA group of studies have demonstrated eight different types of data segment techniques including migration [56, 57, 105, 111, 119 ], shadow data structure [59, 74, 105 ], copy field values [41 ], swapping [95 ], type wrapping [105 ], screen buffering and data caching [73 ], Struct replacement [22 ], and shared data structure [68 ]. C.…”

“…A number of investigations have presented five types of code segment techniques including indirection [4, 46, 47, 53, 55, 74, 77, 94, 99, 106, 119 ], injection [46, 74, 79, 91, 103, 120 ], rewriting [47, 50, 53, 74, 106 ], bytecode manipulation [41, 43, 55 ], and trampolines [4, 106 ].…”

“…Furthermore, the number of investigations have illustrated extra application domains including onboard software attitude orbit control [110 ], power plant controller system [63 ], high‐performance computing [47 ], braking applications [133 ], data destitution centric applications [121 ], scientific applications [50 ], intelligent field devices [98 ], self‐adaptive systems [91 ], safety‐critical embedded systems [53 ], critical software‐intensive systems [122 ], message‐passing programs [118 ], transportation or space applications [122 ], dynamic vehicle software [66 ], speed up automatic program repair [30 ], and enhance security and privacy in high availability energy management applications in smart cities [106 ].…”