Hi All, First, thank May for helping present the draft on yesterday. I really appreciate it!
Secondly, I believe most of May's questions raised in her email have been addressed during our offline discussion. Essentially, network information such as link bandwidth can be collected through cost map or ECS, but to provide sufficient information for the scheduler to make optimal scheduling decisions, topology extension services such as path vector, RSA, FCS, etc., are in great need. We plan to deploy RSA on the CMS testbed developed by CalTech for demonstration in SuperComputing 2016. And we will update the details and results on our progress to the working group as soon as possible. Thirdly, to further answer Vijay, Jan and Richard's question, other than multi-domain, another important feature making it different from other network applications is that the data transfer scheduling in science networks need to consider the relation between different flows/datasets. For instance, different datasets may have raw->processed relation, for instance, a processed dataset A may need to wait till other raw datasets transferred to a certain site before A can be transferred. In current development, we have not designed scheduling policies for supporting general dataset transferrs relations. We plan to explore an efficient scheduling algorithm for this scenario in the short term future. Thanks. Best Qiao On Sat, Jul 16, 2016 at 10:09 AM, May （Haoran） <[email protected] <javascript:_e(%7B%7D,'cvml','[email protected]');>> wrote: > Dear Qiao and all, > > This draft is a great example to provide a scenario for using ALTO > information. However, I have several questions and want to make it clear: > > First of all, if I understand right, EDTO has three information sources: > the DTR, the network information from ALTO, and the information of the > datasets. When you mentioned that ALTO server will provide link bandwidth > in line 3 in section 5.4, I’m not very clear why and how ALTO server > provides this kind of information. > > Besides, if EDTO will do the scheduling work, supposedly it will firstly > get the information of link capacity. But in the architecture of EDTO now, > I’m not quite sure how does the scheduler get the information of link > capacity. > > One small suggestion, the diagram in section 5.1 which shows the > architecture of EDTO platform, will be more precise if it can display the > direction of dataflow. For example, the data from users’ requests will flow > to the DTR Collector, and the data from DTR Collector, Dataset Transfer > Manager and ALTO Client will flow to DTR Scheduler, etc. Using arrows to > replace the lines when there is dataflow will be more clear and > understandable. > > Plus, three trivial problems. First is that the page links of 5.7.3 and > 6.4 doesn’t work. Second, in Section 1, Paragraph 2, line 4, ANd -> And. > Third, in Page 4, first line, not sure if you want to say flow steering by > saying flow sterring( a typo?) > > Please correct me if I misunderstand anything. Thanks! > > May > > At 2016-07-09 07:53:04, "Qiao Xiang" <[email protected] > <javascript:_e(%7B%7D,'cvml','[email protected]');>> wrote: > > Dear All, > > We have submitted a new I-D for consideration. It describes a dataset > transfer orchestrator framework for exascale science networks. It collects > network information from multiple ALTO services utilizing proposed > topology extensions and > leverages emerging SDN control capabilities to orchestrate the scheduling > of multiple large dataset transfers, leading to improved data transfer > latency and reliability as well as more efficient utilization of limited > network resources. > > It can be found at > https://datatracker.ietf.org/doc/draft-xiang-alto-exascale-network-optimization/ > . and your comments are welcomed. > > Thanks. > > > > Best > Qiao > -- > Qiao Xiang > Postdoctoral Fellow, > Department of Computer Science, > Yale University > > > > > -- Qiao Xiang Postdoctoral Fellow, Department of Computer Science, Yale University -- Qiao Xiang Postdoctoral Fellow, Department of Computer Science, Yale University
_______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
