Optimizing Large Data Transfers over 100Gbps Wide Area Networks

Rajendran, Anupam; Mhashilkar, Parag; Kim, Hyun Woo; Dykstra, Dave; Garzoglio, Gabriele; Raicu, Ioan

doi:10.1109/ccgrid.2013.25

Cited by 8 publications

(9 citation statements)

References 11 publications

(9 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…U : Utilization L : Number of processes issuing requests p : Probability that a request will access a given disk γ : Empirically calibrated value (1) The disk throughput can be determined by Equation (2) in which the disk utilization in Equation 1 is multiplied by N ·SU E(S) . The equation can be rearranged as Equation (3) after…”

Section: ) Disk Modeling: Disk Edgementioning

confidence: 99%

“…saturate a 100G network link. The results in [1] show that they can achieve 80-90 Gbps in case of memory-to-memory data transfer, where the system's RAM buffer cache is big enough to hold the entire dataset, so the dataset is loaded into memory before data transfer.…”

Section: Introductionmentioning

confidence: 96%

“…In [1], various data transfer middleware such as GridFTP [2] and SRM [3] has been evaluated to determine whether they can…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Toward optimizing disk-to-disk transfer on 100G networks

Jung

Kettimuthu

Vishwanath

2013

2013 IEEE International Conference on Advanced Networks and Telecommunications Systems (ANTS)

View full text Add to dashboard Cite

The recent emergence of ultra high-speed networks up to 100 Gbps has posed numerous challenges and has led to many investigations on efficient protocols to saturate 100 Gbps links. However, end-to-end data transfers involve many components, not only protocols, affecting overall transfer performance. These components include a disk I/O subsystem, additional computation associated with data streams, and network adapter capacities. For example, achievable bandwidth by TCP may not be implementable if disk I/O or CPU becomes a bottleneck in end-to-end data transfer. In this paper, we first model all the system components involved in end-to-end data transfer as a graph. We then formulate the problem whose goal is to achieve maximum data transfer throughput using parallel data flows. Our contributions lie in how to optimize data transfers considering all the system components involved rather than in accurately modeling all the system components involved. Our proposed formulations and solutions are evaluated through experiments on the ESnet 100G testbed. The experimental results show that our approach is several times faster than Globus Online -8x faster for datasets with many 10MB files and 3-4x faster for other datasets of larger size files.

show abstract

Section: ) Disk Modeling: Disk Edgementioning

confidence: 99%

Section: Introductionmentioning

confidence: 96%

See 1 more Smart Citation

Toward optimizing disk-to-disk transfer on 100G networks

Jung

Kettimuthu

Vishwanath

2013

2013 IEEE International Conference on Advanced Networks and Telecommunications Systems (ANTS)

View full text Add to dashboard Cite

show abstract

“…conducted on new 100G high-speed networks. In [1], various data transfer middleware such as GridFTP [2] and SRM [3] has been evaluated to determine whether they can saturate a 100G network link. The results in [1] show that they can achieve 80-90 Gb/s in case of memory-to-memory data transfer, where the system's RAM buffer cache is big enough to hold the entire dataset, so the dataset is loaded into memory before data transfer.…”

mentioning

confidence: 99%

Cluster-to-cluster data transfer with data compression over wide-area networks

Jung

Kettimuthu

Vishwanath

2015

Journal of Parallel and Distributed Computing

View full text Add to dashboard Cite

h i g h l i g h t s• We model all the system components involved in end-to-end data transfer as a graph.• We formulate the problem to optimize data transfer throughput using parallel flows.• We propose a novel software I/O stack with variable parallel data flows per layer. • Optimized parallel data flows improve overall data transfer throughput.• The novel I/O stack with data compression alleviates network bottleneck situations. a b s t r a c tThe recent emergence of ultra high-speed networks up to 100 Gb/s has posed numerous challenges and has led to many investigations on efficient protocols to saturate 100 Gb/s links. However, end-toend data transfers involve many components, not only protocols, affecting overall transfer performance. These components include disk I/O subsystem, additional computation associated with data streams, and network adapters. For example, achievable bandwidth by TCP may not be implementable if disk I/O or CPU becomes a bottleneck in end-to-end data transfer. In this paper, we first model all the system components involved in end-to-end data transfer as a graph. We then formulate the problem whose goal is to achieve maximum data transfer throughput using parallel data flows. We also propose a variable data flow GridFTP XIO stack to improve data transfer with data compression. Our contributions lie in how to optimize data transfers considering all the system components involved rather than in accurately modeling all the system components involved. Our proposed formulations and solutions are evaluated through experiments on the ESnet 100G testbed and a wide-area cluster-to-cluster testbed. The experimental results on the ESnet 100G testbed show that our approach is several times faster than Globus Online-8 × faster for datasets with many 10 MB files and 3-4 × faster for other datasets of larger size files. The experimental results on the cluster-to-cluster testbed show that our variable data flow approach is up to 4 × faster than a normal cluster data transfer.

show abstract

“…These solutions tune TCP and other networking-stack parameters at senders and receivers, use parallel TCP connections, stripe transfers using multiple servers at each end with parallel file systems [24,25], and pipelining, which is especially useful if a dataset has many small files [26,27]. Other papers have proposed the use of multi-hop path splitting [28].…”

Section: Related Workmentioning

confidence: 99%

A Cross Layer Design for Large Data Transfers

Alali¹

View full text Add to dashboard Cite

Software Defined Network (SDN) technologies have enabled the introduction of new services such as dynamic Layer-1 (L1) circuits and Layer-2 (L2) virtual circuits (VCs). Our objective is to fully realize high-rate circuits/VCs with almost zero packet loss for large dataset transfers. A multidomain SDN dynamic Layer-2 (L2) VC was used to conduct experiments to achieve this objective.The results showed that a combination of (i) Circuit TCP (CTCP), in which the sending rate is held fixed, and (ii) Token Bucket Filter (TBF) rate shaper at the sending host, is best to achieve, high-throughput transfers. However, packet losses were still observed on these L2 VCs. Therefore, a new study was presented using more controlled environments.A new experimental study was undertaken in a single-rack testbed in which packet losses and delays could be deliberately controlled. Three cases were emulated: (i) single circuit/rateguaranteed VC for a single large transfer from a server, (ii) multiple simultaneous large transfers from a server, and (iii) semi-rate-guaranteed VC. CTCP and the TBF queueing discipline of the Linux traffic control (tc) utility are recommended for the first case, and parameter selection methods are provided. For the second case, the tc Hierarchical Token Bucket (HTB) discipline is required to support multiple transfers, each on a distinct VC with its own rate and Round-Trip Time (RTT). Our experiments showed that dynamic additions and deletions of classes are possible without impact on ongoing large-transfer flows. For the third case, CTCP is recommended if the throughput of the large transfer is of primary concern, while HTCP is recommended if higher consideration should be given to the other flows.The thesis also provides insight and lessons learned about the to Linux TCP/IP stack. Three layers were considered: (i) Application layer, (ii) Transport layer, and (iii) Data-link layer. New iv v findings, which are beneficial to networking researchers are presented. We also provide insights into how to monitor flows using tools such as tcpdump, tshark, and tcptrace. AcknowledgmentsI would like to take this opportunity to thank many people without whom this thesis would have not

show abstract

Optimizing Large Data Transfers over 100Gbps Wide Area Networks

Cited by 8 publications

References 11 publications

Toward optimizing disk-to-disk transfer on 100G networks

Toward optimizing disk-to-disk transfer on 100G networks

Cluster-to-cluster data transfer with data compression over wide-area networks

A Cross Layer Design for Large Data Transfers

Contact Info

Product

Resources

About