The "memory wall", first predicted in the mid 1990's in a paper titled "Hitting the Memory Wall: Implications of the Obvious", was supposed to have emerged by now. However, the typical application instance still only utilizes 5-10% of a modern server's capacity, and the percentage is dropping. In this Webinar, Gil Tene (CTO, Azul Systems) will introduce the notion of the Application Memory Wall, and discuss the dominant factors limiting the scale and performance of modern server application instances. Gil will show how GC behavior has come to dominate application scale and architecture decisions, highlight the resulting constraints and challenges in effectively utilizing current- generation servers, and discuss innovations that may make the 'Application Memory Wall' disappear in the near future.
Title: Java GC and the Application Memory Wall
Speaker: Gil Tene, VP of Technology, CTO and Co-Founder, Azul Systems
The “memory wall”, first predicted in the mid-1990s in a paper titled “Hitting the Memory Wall: Implications of the Obvious”, was supposed to have emerged by now. As CPU throughput increased driven by Moore’s Law, the relatively slow improvements in memory bandwidth and memory latency were expected to have reached a level of disparity in which virtually all applications would be memory bandwidth and/or latency bottlenecked some time in the previous decade.
Yet the current reality for server applications is that It now takes many application instances running simultaneously to even come close to challenging the abundant memory capacity, CPU throughput, and bandwidth available in commodity servers.
So, were all those “hitting the XYZ wall” predictions flat-out wrong? Or is there something else, an even bigger scalability wall that was quietly hit sometime in the past decade that has kept us from running into those other walls? For anyone who has experienced a multi-second Garbage Collection pause in the recent past, the answer may be intuitively obvious…
In this presentation, Gil Tene (CTO, Azul Systems) introduces the notion of the Application Memory Wall and discusses the dominant factors limiting the scale and performance of modern server application instances. Gil shows how GC behavior has come to dominate application scale and architecture decisions, highlights the resulting constraints and challenges in effectively utilizing current-generation servers, and discusses innovations that may make the ‘Application Memory Wall’ disappear in the near future.