On the suitability of using circuit-switched networks for storage applications

We identified three storage applications, disaster-recovery (DR) for critical data, backup-storage for ordinary users’ data, and server-replication of web servers and email servers for a quick recovery from failures, as potential candidates for dynamically shared circuit-switched networks. We experimented with a popular, open-sourced backup-storage application, called amanda, and found no technical impediments to adapting this code to work on dynamically shared circuitswitched networks. Further, we obtained information about backup sizes and frequencies of backups from a survey of amanda users. This data shows that typical data transfers are quite large, which makes the call-setup delay overhead a small fraction of the total call-holding time. This means it feasible to use dynamically circuit-switched networks for this application. As a second, somewhat independent, thread of the work, we conducted our own survey of regional and enterprise research-and-education (R&E) networks for their use of storage applications, with a goal of understanding where dynamically shared circuit-switched networks should be deployed to support these applications. Based on the responses, we conclude that traffic generated by storage applications is primarily intra-regional, with the exception that some enterprises purchase these types of storage services, especially DR, from commercial Application Service Providers (ASPs). ASPs typically locate their servers in a few PoPs, which implies that in such a usage, storage traffic will flow across core (backbone) networks in addition to traversing enterprise and regional networks. In the latter case, core Dynamic Circuit Services (DCS) networks, such as the one deployed by Internet2, can be used in a hybrid mode with circuits across the core network concatenated to TCP connections over IP-routed regional/enterprise networks. Furthermore, we learned that since data movement for DR and server-replication applications is limited to a few transfers (just critical data for DR and a few servers for server-replication), the additional burden posed to the IP network is minimal, which allows for most enterprises to simply use their IP-routed networks for this traffic. Backup-storage of ordinary users’ files (through applications such as amanda) was not supported by any of the regional or enterprise IT network managers who responded to our survey. This application is likely to require much higher levels of aggregate bandwidth, given the larger number of users. At these higher rates, it becomes more feasible to economically justify using dynamically shared circuit-switched networks instead of simply overprovisioning IP-routed networks. Thus, we see the potential for a business case for deploying high-speed DCS networks within regional and enterprise networks if backup-storage service is offered from servers in regional or core network PoPs.