Autonomic systems. Most autonomic systems projects study individual autonomic properties, specific self- managed systems, or focus on specific elements of autonomic behavior. Little research has considered the overall architectural implications of building self-managed distributed systems. The SELFMAN project is unique in this respect, combining as it does component-based system construction with overlay network technology into a service architecture for large-scale distributed system self management. The focus on both large-scale, loosely coupled systems, and architectural issues is a key differentiator of the SELFMAN project with respect to most of autonomic systems research. We now present each of these areas in more detail and explain where the contribution of SELFMAN fits in. The RAD Laboratory (Reliable, Adaptive, Distributed systems) was recently created at UC Berkeley and is funded by Google, Microsoft, and Sun. XXX intends to use research from statistical learning theory, control theory, and machine learning to improve the detection of problems in distributed systems, and to apply research from Recovery-Oriented Computing to provide fast recovery and reaction mechanisms that tie in to this. As such, XXX’s vision overlaps with the SELFMAN vision. RAD proposes a specific solution path for improving reliability and adaptability for distributed systems. We consider that RAD and SELFMAN are complementary: RAD has a broad coverage of issues related to the construction of distributed applications, including considerations for Internet architecture and network monitoring. Also, RAD places a strong emphasis on statistical learning tools for assessing the behavior of distributed applications and integrating those in development tools and middleware. In contrast, SELFMAN has a more restricted scope, with little considerations for behavior assessment and diagnostics. SELFMAN will closely follow the RAD work and use its ideas where possible.
Autonomic systems. The main goal of autonomic system research is to automate the traditional functions associated with systems management, namely configuration management, fault management, performance management, security management and cost management [DMTF]. This goal is becoming of utmost importance because of increasing system complexity. It is this very realization that prompted major computer and software vendors to launch major R&D initiatives on this theme, notably, IBM's Autonomic Computing initiative and Microsoft's Dynamic Systems initiative. The motivation for autonomic systems research is that networked environments today have reached a level of complexity and heterogeneity that make their control and management by human administrators more and more difficult. The complexity of individual elements (a single software element can literally have thousands of configuration parameters), combined with the brittleness inherent of today's distributed applications, makes it more and more difficult to entertain the presence of a human administrator in the “management loop”. Consider for instance the following rough figures [GANE03]: - One-third to one-half of a company's total IT budget is spent preventing or recovering from crashes. - For every dollar used to purchase information storage, 9 dollars are spent to manage it. - 40% of computer system outages are caused by human operator errors, not because they are poorly trained or do not have the right capabilities, but because of the complexities of today's computer systems. IBM's autonomic computing initiative [IBM], for instance, was introduced in 2001 and presented as a “grand challenge” calling for a wide collaboration towards the development of computing systems that would have the following characteristics: self configuring, self healing, self tuning and self protecting, targeting the automation of the main management functional areas (self healing dealing with responses to failures, self protecting dealing with responses to attacks, self tuning dealing with continuous optimization of performance and operating costs). Since then, many R&D projects have been initiated to deal with autonomic computing aspects or support techniques. For example, we mention the following projects that are most relevant to SELFMAN: - The Recovery-oriented Computing project [ROC] at UC Berkeley, which studies techniques for fault recovery through micro-reboot techniques. - The OceanStore project [OCEANSTORE], also at UC Berkeley, which s...
