230 likes | 243 Views
This research paper explores the Bulk Data Transfer Service (BDTS) and its evaluation on the IGTMD link. It discusses the motivations for data transmission in a Grid environment, the BDTS architecture, key concepts, flow control mechanisms, estimation techniques, collaboration with existing protocols, and the validation of the system. The paper concludes with open problems and the deployment of BDTS and network resource management in a Grid environment.
E N D
BDTS and Its Evaluation on IGTMD link C. Chen, S. Soudan, M. Pasin, B. Chen, D. Divakaran, P. Primet CC-IN2P3, LIP ENS-Lyon 2008-06-30
Outline • Bulk Data Transfer Service • A framework for Service Differentiation in Grid
Motivations • Data transmission in Grid • Large volume, Long term • Between countable sites • Between storage systems • Between computation nodes and local storage system • Over private shared network • End-to-End quality of service. • Strict deadline • High reliability • Best-Effort with TCP • “Fair sharing” VS. Throughput intensive • ASAP VS. Strict deadline • Burst traffic
BDTS • Bulk Data Transfer Service • A centralized bandwidth allocation system • Provides deadline concerned data movement • End-to-End traffic control • Smooth traffic • Optimizes link utilization
BDTS Architecture • User Interface • Submit Transfer Job (t-job) • Network IS • Provide network information (topology, bandwidth, etc.) • Job Management • Optimization (BDTSh) • Flow control (FLOC) • Control the actual transfer according to the profile from JM
Key Concepts • Network Model • Static layer 2 topology and characters • T-job • Volume: application layer data size • Time: Start time, End time • Source Destination pair • Path from NIS • Max-rate: Limitation due to end systems • Profile • Time-rate: layer 2 rate.
Optimization • Minimize the Congestion Factor (allocated-throughput / Link-bandwidth) • Guarantee full usage of link bandwidth • Reduce the average package delay experienced by the coexisting interactive traffic B. Chen and P. Primet. Schedulling deadline-constrained bulk data transfer to minimize network congestion. In CCGRID’07, Pages 410-417. May 2007
Flow Control • TCP + Precise Software Pacer (PSPacer) • PSP • A module for iproute2 • Precise network bandwidth control • Traffic Shaper • TCP with congestion avoid mechanism disabled • Optimized in restramssion and memory management • Collaborating with existing TCP based application
The Estimation • No Ideal Transport Protocol: Layer 2 profile VS. Layer 7 Date Transfer • Protocol overhead (L7 protocol, TCP/IP header) • Synchronization between the profile and implementation • TCP’s re-act to the network • … • Linear estimation: Vs = a * Vr + b
The Project • Job Management (Java, c++) • User Interface (Java) • Floc / API (C, Linux Kernel, PSP) • Gridftp client (c++, globus)
On Grid5000 • GridFTP • Security • File transfer
Long latency Average 106ms Background Traffic IGTMD
Conclusions • Conclusions • Bandwidth resource management based on End-to-End traffic control • Provide a deadline file transfer based on IP • Open problems • The effect of arriving time of t-jobs to the optimization • The efficiency of Flow control • The deployment of BDTS and network resource management in Grid
Outline • Bulk Data Transfer Service • A framework for the Service Differentiation in Grid
Grid-Managed Network Resources • Why the network (bandwidth/Qos) need to be managed in a Grid running on a dedicated private Giga/Tera network? • More bandwidth, more bandwidth consuming applications • each member will have its own demand into the network. • Grid network • Applications sharing common infrastructure • Middle-ware service • Hosts, storage systems, networks, • Network is transparent for Applications • Introduce differentiation of service to application level objects?
How? • Static • Different quality of service is provided by different instances • Applications make a choice • Dynamic • Service identify application • Assign SLA to instance of applications • Knowledge of both its users and its underlying service
A “ulimit” like network resource management interface • Adapt networking policy to traffic according to the process it belongs to • Identify • Traffics from a process and its children • Traffics from a user • Traffics from a VO • Traffics from a certain application • Traffics to certain destinations • Policy • Output Bandwidth • Diffserv QoS • … • Running time adjusting • Inner adjusting: • Grid-Job Wrapper: Deployment-User application-Uploading Result • Outer adjusting • Grid network Resource Management • Wrapper
Thank you! • Comments and Questions?