Alan Bivens

Architectural design for next generation heterogeneous memory systems

New enterprise workloads requiring fast, reliable access to increasing amounts of data have pushed todays memory systems to power and capacity limits while creating bottlenecks as they ensure transactions are persistently tracked for reliability. New storage class memory technologies (such as phase change memory) have the potential to offer high capacity within latency and bandwidth ranges acceptable for a computer memory system and persistence which may help ease the system-level burden of balancing performance and reliability. This paper describes architectural options for addressing the challenges of future, heterogeneous memory systems as well as the attributes required of the next generation memory devices.

Comparing the use of bayesian networks and neural networks in response time modeling for service-oriented systems

The new paradigm of service-oriented computing facilitates easy construction of dynamic, complex distributed systems. Recent research has shown that machine learning methods can be a promising way to autonomously and accurately derive models to assist autonomic management software or humans in understanding system behaviors and making informed decisions. However, the efficacy of different machine learning techniques in describing various system behaviors and meeting distinct application needs has not been systematically understood. Such an understanding can prove crucial in management infrastructure design and implementation for service-oriented systems.n This paper is an initial step to bridge the gap and specifically contrasts the applications of Bayesian networks (BN) and neural networks (NN) in modeling the response time of service-oriented systems. Relatively simple BN and NN models are designed and implemented as a base of the comparison study. As far as model performance is concerned, a wide range of simulations show that BNs offer better accuracy, are less sensitive to small data set size and are therefore more suited for environments that change rapidly and need frequent response time model reconstructions; whereas NNs can achieve faster model evaluation time and support management routines that demand intensive response time predictions. From a non-performance perspective, it is analytically concluded that BNs can be more easily understood by human and support multi-direction evaluation, while NNs provide more flexible response time representation.

Adaptive task duplication using on-line bottleneck detection for streaming applications

In this paper we describe an approach to dynamically improve the progress of streaming applications on SMP multi-core systems. We show that run-time task duplication is an effective method for maximizing application throughput in face of changes in available computing resources. Such changes can not be fully handled by static optimizations. We derive a theoretical performance model to identify tasks in need of more computing resources. We propose two on-line algorithms that use indications from the performance model to detect computation bottlenecks. In these algorithms, a task can identify itself as a bottleneck using only its local data. The proposed technique is transparent to end programmers and portable to systems with fair scheduling. Our on-line detection algorithms can be applied to other dynamic scenarios, for example, involving run-time variation of workload.n Our experiments using the StreamIt benchmarks [5] show that the proposed run-time task duplication achieves considerable speedups over the multi-threaded baseline on a 16-core machine and on the scenarios with dynamically changing number of processing cores. We also show that our algorithms achieve better application throughput than alternative approaches for task duplication.

Performance problem localization in self-healing, service-oriented systems using Bayesian networks

In distributed, service-oriented environments, performance problem localization is required to provide self-healing capabilities and deliver the desired quality of service (QoS). This paper presents an automated approach to identifying system elements causing performance problems. Applying probabilistic inference to collected response time and elapsed time data, the approach 1) infers elapsed time for services where data is missing, 2) estimates the response time degradation caused by different services using the duration, abnormality and response time correlation of their elapsed times, and 3) identifies the services that are the most important causes of slow response time and yield the most benefit if recovered. The approach has been used to localize a performance problem on the test bed of a real-world service-oriented Grid. Evaluation using simulations shows that the approach consistently achieves better accuracy than traditional techniques in various service-oriented settings.

Performance of large low-associativity caches

While it is known that lowering the associativity of caches degrades cache performance, little is understood about the degree of this effect or how to lessen the effect, especially in very large caches. Most existing works on cache performance are simulation or emulation based and there is a lack of analytical models characterizing performance in terms of different configuration parameters such as line size, cache size, associativity and workload specific parameters. We develop analytical models to study performance of large cache architectures by capturing the dependence of miss ratio on associativity and other configuration parameters. While high associativity may decrease cache misses, for very large caches the corresponding increase in hardware cost and power may be significant. We use our models as well as simulation to study different proposals for reducing misses in low associativity caches, specifically, address space randomization and victim caches. Our analysis provides specific detail on the impact of these proposals, and a clearer understanding of why they do or do not work.

Program behavior characterization in large memory systems

We introduce models to characterize large cache performance in terms of various statistics related to sojourn time of a line in the cache. These statistics themselves depend on cache configuration parameters and we are currently working to isolate this dependency using LCS data and models. This will then help us in obtaining explicit relation between cache performance and its configuration parameters which will be helpful in identifying an optimal set of configuration parameters during early design phase of large memory systems.

Efficient Statistical Performance Modeling for Autonomic, Service-Oriented Systems

As service-oriented environments grow in size and complexity, managing their performance becomes increasingly difficult. To assist administrators, autonomic techniques have been adopted to permit these environments to be self-managing (problem localization, workload management, etc.). These techniques need a sense of system state and the ability to project a new state given some change within the environment. Recent work addressing this issue frequently used statistically learned models which were derived entirely from data. However, many environments already have management facilities in place that could provide precise and useful insights (e.g. workflows) into the system. This paper introduces a method of modeling service-oriented system performance using Bayesian networks and specifically addresses the benefits obtained by incorporating these insights into the model learning process. To further minimize model building costs, we devise a decentralized method to concurrently learn parts of the model where knowledge inclusion is impossible. Simulations and applications in actual environments show significant reductions in learning time, better accuracy and stronger tolerance to small learning data sets.

Performance modeling and characterization of large last level caches

For each different cache configuration, we identified parametric functions to characterize the probability density of statistics related to cache performance. Observe that, the parameters (α,β) in (1) and (a, b, n) in (2) depend on cache configuration. We next need to study the sensitivity of the parameters in the density function to cache configuration parameters. Once this is established we can express the distribution of CRT, SRT and IHT in terms of cache size and line size alongwith workload specific constants. Having known these distributions, T crt , T srt and T iht and hence m from Model-A and Model-B can be written as an explicit function of cache size and line size for a given workload.

Advanced search system for IT support services

IBMs Technical Support Services division runs remote support centers, where agents provide phone support for client problems related to IBM and non-IBM hardware and software products. Support center personnel use numerous pieces of information—including many searches, log files, and records of historical support tickets, from disparate data sources—to recommend solutions for customer technical problems. We have built an advanced search system to assist support agents who are resolving customer service requests and improving our client experience. The system has been deployed and used globally by thousands of support center personnel. In this paper, we describe the systems architecture, the technical challenges, and the innovative solution we have built. In addition, we discuss the novel ideas to address the unique requirements and challenges of the support services domain. These ideas include using system logs and domain knowledge to automatically expand agent queries, incorporating implicit agent feedback, and selecting features to extract useful information from highly unstructured and noisy ticket data. Results on the effectiveness of the system are presented. We also discuss future work on enhancing the systems capability to automatically diagnose customer hardware and software problems and remediate them.

Towards More Effective Solution Retrieval in IT Support Services Using Systems Log

Technical support agents working in the IT support services field resolve IT problems. They are often faced with the daunting task of identifying the correct solution document through a search system from large corpora of IT support documents. Based on the observation that system logs may contain critical information for identifying the root cause of IT problems, we explore the idea of automatic query expansion by using system logs as a bridge to link queries with the most relevant documents. Given the original query from a user such as a technical support agent, an intermediate query is first formed by adding key terms extracted from system logs using domain-specific rules. Based on topic models, further key terms are selected from corpora of IT support documents, which are combined with the intermediate query to form the final query. Our experimental results show that expanding queries using system logs together with topic models yields better performance in retrieving relevant IT support documents than using topic models only.