<html><head><meta http-equiv="Content-Type" content="text/html charset=utf-8"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;" class="">This is a reminder about Sam Nickolay’s MS Presentation on Monday morning.<div class=""><br class=""></div><div class="">——————————</div><div class=""><div class="">Department of Computer Science</div><div class="">The University of Chicago</div><div class=""><br class=""></div><div class="">Date: Monday, November 21, 2016</div><div class=""><br class=""></div><div class="">Time: 9:00 am</div><div class=""><br class=""></div><div class="">Place: Ryerson 277</div><div class=""><br class=""></div><div class="">Bx/MS Candidate: Sam Nickolay</div><div class=""><br class=""></div><div class="">MS Paper Title: Towards Bridging the Gap between Peak and Average Loads on Science Networks</div><div class=""><br class=""></div><div class="">Abstract:</div><div class="">Backbone networks are typically overprovisioned in order to sustain peak<br class="">loads. Research and education networks (RENs) are designed to operate at<br class="">20-30% loads. For example, Internet2 upgrades the backbone interconnects<br class="">when the weekly 95th-percentile load is reliably above 30% of link<br class="">capacity. Our analyses of traffic on ESnet between DOE facilities also<br class="">show that there is a huge gap between peak and average utilization. As<br class="">science data volumes increase exponentially, it is unclear whether this<br class="">over provisioning trend will continue in the future. Even if over<br class="">provisioning is possible, it may not be the most cost-effective (and<br class="">desirable) approach going forward. Under the current mode of free access<br class="">to RENs, the traffic at peak load may range from flows that need to be<br class="">transferred in near-real time, for example,<br class="">for computation and instrument monitoring and steering, to flows that<br class="">are less time-critical, for example, archival and storage replication<br class="">operations. Thus, peak load does not necessarily indicate the capacity<br class="">that is absolutely required. In this paper, we study how data transfers<br class="">are impacted when the average network load is increased while the<br class="">network capacity is kept at the current levels. We also classify data<br class="">transfers into on-demand (ones that are time-critical) and best-effort<br class="">(ones that are less time-critical) and study the impact on both classes<br class="">for different proportions of both the number of on-demand transfers and<br class="">amount of bandwidth allocated for on-demand transfers. We use real<br class="">transfer logs from production GridFTP servers for our study. Our results<br class="">indicate that the average slowdown experienced by the data transfers is<br class="">under 1.5x even when the load is doubled with the network capacity fixed<br class="">at the current levels. When the transfers are classified into on-demand<br class="">and best-effort, on-demand transfers experience almost no slowdown and<br class="">the slowdown experienced by best-effort transfers is under 2x, even when<br class="">50% of transfers were treated as on-demand.</div><div class=""><br class=""></div><div class="">Sam’s MS advisor: Ian Foster</div><div class=""><br class=""></div><div class="">A draft copy of the MS paper is attached.</div><div class=""><br class=""></div><div class=""></div></div></body></html>