Large enterprises connect their locations together by building a corporate network out of private communication channels, such as leased lines, frame relay and ATM links, called physical private networks or PPNs. PPNs provide good quality of service, but they are expensive. On the other hand, Internet-based virtual private networks (VPNs) can provide speedy deployment of multisite corporate networks at a small fraction of the cost of private lines. However, Internet-based VPNs do not offer the same accountability and predictability as PPNs do, since the Internet is not administrated by a single provider. In order to build a hybrid network that is both reliable and affordable, we have developed the delta routing protocol, which allows nodes on a corporate network to communicate with each other using both PPN and Internet-based VPN. The delta routing protocol works with the existing routing protocol on the PPN and allows each node to determine the best routes on the hybrid network using local information only. We have simulated and compared the delta routing protocol and the alternatives using the ns-2 simulator. The results show that the delta routing protocol outperforms the alternatives in a variety of scenarios.

In a multipath network with a naive traffic partitioning scheme, varied packet loss rates as well as varied delays on multiple paths could seriously reduce TCP bandwidth. We propose a scheme, called dial-controlled hash (DCH), for dynamically partitioning traffic flows across multiple paths. DCH allows for fine-grained load balance, minimizes path oscillation for each traffic flow, and does not require per-flow state maintenance or packet tagging. We evaluate DCH in two simulated networks, a parallel link topology and a corporate intranet with trace-based traffic generation. In both simulations, we compare the proposed scheme to a number of alternatives. DCH reduces path oscillation by a factor of 1.6-37 and improves TCP bandwidth by up to 66that, without a good flow partitioning scheme, there is a limit on how much one can save by using low-quality networks in combination with high-quality ones.

We study management strategies for main memory database clusters that are interposed between Internet applications and back-end databases as content caches. The task of management is to allocate data across individual cache databases and to route queries to the appropriate databases for execution. The goal is to maximize effective cache capacity and to minimize synchronization cost. We propose an affinity-based management system for main memory database cLUsters ( ALBUM ). ALBUM executes each query in two stages in order to take advantage of the query affinity that is observed in a wide range of applications. We evaluate the data/query distribution strategy in ALBUM with a set of trace-based simulations. The results show that ALBUM reduces cache miss ratio by a factor of 1.7 to 9 over alternative strategies. We have implemented a prototype of ALBUM, and compare its performance to that of an existing infrastructure: a fully replicated database with large buffer cache. The results show that ALBUM outperforms the existing infrastructure with the same number of server machines by a factor of 2 to 7, and that ALBUM with only 1/3 to 1/2 of the server machines achieves the same throughput as the existing infrastructure.

Multithreaded programming is an effective way to exploit concurrency, but it is difficult to debug and tune a highly threaded program. This paper describes a performance tool called Tmon for monitoring, analyzing and tuning the performance of multithreaded programs. The performance tool has two novel features: it uses "thread waiting time" as a measure and constructs thread waiting graphs to show thread dependencies and thus performance bottlenecks, and it identifies "semi-busy-waiting" points where CPU cycles are wasted in condition checking and context switching. We have implemented the Tmon tool and, as a case study, we have used it to measure and tune a heavily threaded file system. We used four workloads to tune different aspects of the file system. We were able to improve the file system bandwidth and throughput significantly. In one case, we were able to improve the bandwidth by two orders of magnitude.