Title: Collaborative research: End-to-End Provisioned Optical Network Testbed for Large-Scale escience Applications

Transcription

1 Year 3 Activities report for the NSF project EIN Title: Collaborative research: End-to-End Provisioned Optical Network Testbed for Large-Scale escience Applications Date: July 21, 2006 PI: Malathi Veeraraghavan, mv5g@virginia.edu Please reiterate the goals and objectives of your efforts, and summarize the research and education activities you have engaged in that aim to achieve these objectives. Include experiments you have conducted, the simulations you have run, the collecting you have done, the observations you have made, the materials you have developed, and major presentations you have made about your efforts. In a later section you will list more formally any publications and other specific products (database, collections, software, inventions, etc.) that have resulted. The goal of this project is to develop the infrastructure and networking technologies to support a broad class of escience projects and specifically the Terascale Supernova Initiative. Our objectives are to design and deploy a high-performance, experimental optical network infrastructure and to test application/ middleware/transport protocol software, developed specifically for escience projects, on this network. Our two target applications are file transfers and remote visualization. The objective of our basic research activities this year has been centered around comparing the conditions under which the Immediate-Request (IR) and Book-Ahead (BA) modes of bandwidth sharing are suitable. We have developed analytical models and run simulations to understand the fundamental aspects of these two bandwidth-sharing modes. The objective of our experimental research activities has been to extend the CHEETAH network to connect University of Virginia (UVa) and City University of New York (CUNY), and complete release of a CHEETAH software package for Linux end hosts, consisting of the circuit requestor, which allows a user to request dedicated 1Gb/s Ethernet circuits to remote hosts whenever needed, and Circuit-TCP (CTCP) software to provide transport-layer functions across dedicated endto-end circuits. The objective of our educational activities was to graduate some of our Masters and doctoral students supported by this grant, and offer our undergraduate students opportunities to conduct experiments and write software for the CHEETAH network. More generally, our goal was to have students learn about optical networks, and acquire inter-disciplinary skills spanning operating systems, distributed systems, parallel computing and networking. 1

2 Below is a summary list of all our activities in this project to date (Aug July 2006), which includes experiments we have conducted, and some observations we have made (more details on our observations are listed in the findings attachment): CHEETAH wide-area network (primarily work of postdoctoral fellow, Xuan Zheng, with support from graduate student, Xiangfei Zhu): Fig. 1 shows the current CHEETAH network. This year, we extended the reach of the CHEETAH network to UVa and CUNY through the Figure 1: Current CHEETAH network deployment (as of July 21, 2005) use of Vortex links in Virginia, NYSERnet connectivity in New York, a HOPI VLAN from New York to Washington, two MPLS tunnels from Washington to Raleigh, NC on the Abilene and NCREN networks. With this connectivity, we are able to connect our laboratories at UVa and CUNY via 1GbE links to the CHEETAH network. Details of this connection are shown in Fig. 2. The second enhancement was that ORNL provided us with the OC192 link between the Sycamore switches at Atlanta and ORNL. Last year we were using two 1Gb/s MPLS tunnels between these switches. Next, we added Georgia Tech to the CHEETAH network. Prof. Karsten Schwan, CS department, Georgia Tech, is planning experiments on this network. Next, we completed implementation of a secure control-plane network for CHEETAH, designed for scalability, and completed a thorough document on our design ( Finally, we have deployed a centralized name server for the CHEETAH network and configured DNS data (including the TXT 2

3 resource record) for all CHEETAH hosts. This has been working in a stable manner. Soon we will distribute this data to operational DNS servers in each enterprise where CHEETAH hosts are located. We published a paper on the CHEETAH testbed in IEEE Comm. Mag., Aug issue. Figure 2: The 1GbEthernet links from UVA and CUNY to the CHEETAH Raleigh PoP at MCNC Application support for TSI (primarily work of post-doctoral fellow, Xuan Zheng): We supported Prof. John Blondin in his use of the CHEETAH network for remote visualization using the Ensight software. Data generated from the TSI simulations executed on the ORNL Cray X1 is moved to a 64-node visualization cluster called hawk.ccs.ornl.gov, which runs the Ensight Server-of-Servers (SOS) software and a server on each host of the cluster. As this cluster is required to be located behind the ORNL firewall, we cannot connect it directly to the CHEETAH network. Instead we place a host called zelda4 on the CHEETAH network and run an Ensight software module, referred to as the hub to relay visualization data and control between Ensight clients connected to the CHEETAH network to the Ensight server running on the hawk cluster. The processing capability and memory of host zelda4 is inadequate to run the Ensight SoS and servers; therefore we use the hawk cluster. An Ensight client module is executed on the head node of the viz cluster located at NCSU. This is a 4-node cluster with a head node, called viz-head, connected to three viz-nodes. The viz-nodes run a process called crserver, which is part of the Chromium display software package. The viz- 3

4 Figure 3: Remote visualization across CHEETAH head routes data received by the Ensight client to the various crserver modules to visualize the TSI simulation data on the 6-panel wide-scope display. The high-speed connectivity internal to ORNL (between hawk and zelda4) and the high-speed wide-area circuit across the CHEETAH network help improve the remote visualization experience for the scientist. We tried another configuration in which the hub was executed at the vizhead and zelda4 was configured to merely serve as an IP router forwarding packets from hawk onto the CHEETAH network. With regards to the file transfer application, we demonstrated that once a dedicated GbE circuit is established from ORNL to NCSU, basic ftp over TCP offers excellent throughput (close to 750Mbps disk-to-disk transfers; see our report on documents/others/file-transfers.doc). However, the Cray X1(E), which the TSI scientists use for their computations, has poor network I/O performance. Our ORNL and NCSU co-pis will report on this significant bottleneck. Basic research (work of graduate student, Xiuduan Fang): The goal of this study was to determine what type of applications are well served by GMPLS networks (e.g., CHEETAH), which currently only supports immediate-request (IR) calls. We used two metrics, call blocking probability and utilization, as measures of application suitability. Applications leading to low call blocking probability and high utilization are desirable. We characterized applications by their 4

5 per-circuit bandwidth requirements and mean call-holding time. Then, we divided applications into two types, ones in which the per-circuit bandwidth and mean call-holding time are independent (e.g., remote visualization) and ones in which they are dependent (file transfers) and presented bandwidth sharing models for these two types of applications. In our bandwidth sharing models, we assumed that calls arrive according to a Poisson process and file sizes can be modeled with a Pareto distribution. We also assumed that all calls are of the same type and share bandwidth on a single link. We computed numerical results for both models. Our results are presented in the findings report (item 5). Basic research (work of graduate student, Xiangfei Zhu): We developed a novel discrete-time Markov chain model for book-ahead bandwidth-sharing mechanisms. We used this analytical model and a simulation model to understand the benefits of book-ahead (BA) bandwidth-sharing when compared to the immediate-request (IR) call-blocking mode of bandwidth-sharing in circuit-switched networks. We studied two different BA schemes, BA-all, in which the caller accepts any set of available timeslots, and BA-n, in which the caller specifies n call-initiation time options. Our results are presented in the findings report (item 7). Signaling software (work of graduate student, Xiangfei Zhu): We have completed a stable release of signaling-related software programs for CHEETAH end hosts, which includes the circuit-requestor (with integrated OCS-client), RSVPD and cheetahd daemons. The circuit-requestor offers users an interface through which to request a 1Gb/s Ethernet circuit. It communicates with the cheetahd daemon, which manages the bandwidth of the host s secondary NIC, configures the data-plane tables (IP routing and ARP tables) and communicates with the RSVPD daemon. The latter generates and receives RSVP-TE messages from the Sycamore switch controller. IPsec tunnels are configured through openswan software from each host to its corresponding Sycamore switch PoP to secure the RSVP-TE communication. This end-host software is deployed on the CHEETAH network, and interoperates well with the Sycamore GMPLS implementation. We measured call setup delays across a two-hop path and found it to be 166ms for OC1 or OC3 circuits, and ~5sec for GbEtherent-SONET-GbE Ethernet circuits. This is because the 21 OC1s of the virtually concatenated SONET circuit are set up individually (results are published in an IEEE ICC06 paper). Since this delay is unacceptably high for file transfers, we worked with Sycamore to have them provide us a new release of the signaling software for GbE-SONET-GbE circuits. Using STS-3c as the basic signal type, they reduced the call setup delay for these circuits to around 1.5sec. We have posted the software and a user guide on our cheetah web site. 5

6 Transport protocol (work of graduate student, Anant P. Mudambi): We completed implementation of a Circuit-TCP (CTCP) solution for dedicated high-speed end-to-end circuits. Our results are published in an IEEE ICC06 paper and the software is available on our web site. This solution includes a CTCP patch for the Linux kernel, which modifies both Linux TCP code and Web100 code, and two user-space programs called CTCP-WAD (Work- Around Daemon) and a tester application program. Routing decision software and heterogeneous networks (work of graduate student, Zhanxiang Huang): Our activities centered around creating a measurements-oriented database to collect delay measurements for file transfers on the TCP path. We combined this with an estimate of delay on the circuit path to make a routing decision. The software is now ready for use but has not yet been integrated with our applications. This is the next step. We studied the signaling and routing aspects of interconnecting heterogeneous networks and published this work in an IEEE Comm. Mag. paper in March Project management: We coordinated activities between all four participating organizations. We supported the CUNY team in their development of CVLSR (with design and OSPF testing). We worked with the ORNL and NCSU teams to facilitate the TSI applications of file transfers and remote visualization. We worked closely with ORNL on CHEETAH network enhancements. Educational activities include the following. Three students, Zhanxiang Huang, Xiuduan Fang and Anant P. Mudambi, obtained theirs Masters degrees this year. Xiuduan and Anant wrote MS theses while Zhanxiang obtained an MCS degree. Xiuduan is continuing on with support of this project for a Ph.D. Xiangfei Zhu was advanced to candidacy and will be defending his Ph.D. proposal soon. Undergraduate students, Erik Halseth, Albert Varma, Chris Dahl, Robert Anthony, Mike Wolynec, did their undergraduate senior thesis work on CHEETAH-related projects. Students have taken full advantage of the opportunities provided in realizing this wide-area testbed and learning the use of all the equipment we have procured. Significant learning focussed on the topics of public vs. private IP addressing, control-plane design, writing and testing code to improve robustness, and learning modeling techniques. Materials: The materials we developed include papers and software. The publications are listed in the Products part of this report. We published two journal papers, and three conference papers. Software developed in the project along with presentations are being posted on our project web site: A list of presentations are provided below. 6

7 Presentations (the second and fifth were done by Xuan Zheng, and the rest by Malathi Veeraraghavan): Title of Presentation Meeting Place Date CHEETAH network and applications Remote visualization over the CHEETAH network Applications and CHEETAH Rapidly Reprovisioned Optical Communications Testbed High Quality Video Applications across High-Speed Optical Networks Status of US optical networking R&D Match between Grid Computing and GMPLS networks, Enabling a connection-oriented internet UVa SEAS Research & Technology Showcase at the Fairfax County Economic Development Meeting Meeting of the MCNC Board of Directors MCNC Applications Symposium JFCOM/UVa meeting UVa CS Department Open House NSF/EU Workshop on Optical Networks NSF/EU Workshop on Optical Networks Brookhaven Natl. Labs. Upton, NY UVa Northern Virginia Campus July 13, 2006 RTP, NC April 27, 2006 Research Triangle Park, MC Charlottesville, VA Charlottesville, VA Brussels, Belgium Brussels, Belgium May 16, 2005 April 10, 2006 March 21, 2006 Feb. 25, 2006 June 27-28, 2005 June 27-28, 2005 May 16,