Hi, I would like point out another requirement from HPC. According to DOE target of Exascale computing, the top HPC system will be comprise 100K or 1M nodes. And each node will interconnected with 2TB/s or 250GB/s BW.
http://www.er.doe.gov/ascr/ascac/Meetings/Mar10/AWhite.pdf I am not sure whether the system will choose regular direct interconnection network such as 3-D torus, or irregular indirect network. DCN/DCI requirements clearly meet the later system. Even in case of direct interconnection, I believe some topology finding mechanism will be strongly required for optimal performance and for operation. ---- Katsushi Kobayashi On 2011/05/13, at 10:29, Xu Xiaohu wrote: > Hi Tony and all, > > In the past few years, RRG has done a lot of excellent research work to > explore various ideas and approaches (e.g., map&encap, id/locator split and > translation...) to addressing the Internet routing scalability issue. Today > it seems that the cloud Data Center Network (DCN) and Data Center > Interconnection (DCI) scenarios are facing a similar scalability challenge > (i.e., MAC forwarding table scalability issue). The demand for VM mobility > within the whole large L2 data center network or even across geographically > dispersed data centers is one major driving force of extending the L2 domain > scope larger and larger. > > Although the reason for the MAC forwarding table scalability issue in the > DCN/DCI scenarios is not the same as that for the Internet routing > scalability issue, the ideas and approaches suitable for scaling the > Internet routing system could be utilized to deal with the MAC forwarding > table scalability issue. There are already many such attempts in reality, > especially in the academic circle. VL2 , SEATTLE and MOOSE are good examples > of them. > > Hence I suggest we spend some time to consider whether we could utilize our > experience which was obtained from the past and ongoing Internet routing > scalability solution research to address the similar scalability issue > existed in the DCN/DCI scenarios, for example, we could attempt or even > develop those familiar ideas or approaches mentioned above to address the > MAC table scalability issue in the DCN/DCI scenarios. > > Best regards, > Xiaohu > > >> -----邮件原件----- >> 发件人: [email protected] [mailto:[email protected]] 代表 Tony Li >> 发送时间: 2011年5月9日 1:18 >> 收件人: [email protected] >> 主题: [rrg] Next topic? >> >> >> Hi all, >> >> It's time to start the conversation about where the RRG is headed. > Towards >> that end, I'd like to open the floor for the discussion of topics. Any > _research_ >> topic within the broad area of routing and addressing is appropriate. >> >> I propose the following process: we hold this discussion this month, > trying to >> identify relevant topics. Each topic should have a succinct statement of > its >> goals. At the end of the month, we poll to determine the interest level > in the >> various topics. Those that seem to have critical mass are recommended to >> the IRTF Chair for consideration. >> >> The default answer, if no topic has critical mass, is for the RG to go on > hiatus. >> >> Any questions or comments on the process? >> >> Any proposals of research topics? >> >> The floor is open... >> >> Tony >> >> _______________________________________________ >> rrg mailing list >> [email protected] >> http://www.irtf.org/mailman/listinfo/rrg > > _______________________________________________ > rrg mailing list > [email protected] > http://www.irtf.org/mailman/listinfo/rrg _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
