Profile cover photo
Profile photo
Henry Xu

Post has attachment
Call for Papers: NetworkML 2016

IEEE ICNP Workshop on Machine Learning in Computer Networks (NetworkML

co-located with

24th IEEE International Conference on Network Protocols (ICNP 2016)
November 8-11, 2016, Singapore


Paper submission: July 21, 2016 (23:59 HKT)
Notification of decision: August 21, 2016
Camera-ready: August 28, 2016 (23:59 HKT)

Call for papers

NetworkML 2016 provides a venue for presenting innovative ideas to discuss future research agendas on machine learning in computer networking. We encourage the submission of work-in-progress papers in the areas of applying machine learning for network design, implementation, measurement, management, deployment, as well as implications of computer networks to machine learning algorithms. We look for submissions of previously unpublished work on topics including, but not limited to, the following:

- Protocol design and optimization using machine learning
- Resource allocation for shared/virtualized networks using machine
- Fault-tolerant network protocols using machine learning
- Machine learning aided network management
- Experiences and best-practices using machine learning in operational networks
- Novel security, performance, and monitoring applications using machine learning
- Implications and challenges brought by computer networks to machine learning theory and algorithms
- Data-driven network architecture design
- Data analytics for network information mining
- Deep learning and reinforcement learning in network control
- Learning-based network optimization

Submission Instructions:

Submitted papers must be no longer than 6 pages (US letter size, 10 point font, 12 point leading, 7 inch by 9.25 inch text block) including all content and references. The sig-alternate-10pt.cls style file satisfies the formatting requirements. Compile your source with options that produce letter page size. All submissions must include names and affiliations of all authors on the title page (no anonymization). Papers must contain novel ideas and must differ significantly in content from previously published papers and papers under simultaneous submission.

Submission site:

Program Co-chairs:

Henry Xu, City University of Hong Kong, HK
Baochun Li, University of Toronto, Canada
Yanhui Geng, Huawei Noah’s Ark Lab, HK

Post has shared content
+Luiz André Barroso and I had a great time yesterday giving a tour of Google to a bunch of +Manchester United players and staff. Topics of discussion ranged from whether or not Luiz had the soft Brazilian football touch, to internet balloons, self-driving cars, encryption, deep neural networks and artificial intelligence, and espresso machines. Thanks for visiting, Morgan Schneiderlin, Ander Herrera, Matteo Darmian, Juan Mata, Anders Lindegaard, Daley Blind, Tony Strudwick, John Murtough, Paul Brand and Paulo Gaudino.

Post has attachment

Post has attachment
Bad reviewing of IEEE Network​

We submitted a survey paper, "Low Latency Datacenter Networking", to IEEE Network in August 2014. We received three reviews on April 24, 2015, after 8 months. The verdict is reject. I found the reviewers are not qualified at all as experts working in the area, their comments reckless and irresponsible, to the point that I decide I have to show them here and provide my rebuttal. I am rather disappointed that our paper, and possibly other papers, receive such careless reviewing after a prolonged review cycle.

"Reviewer: 2

Comments to the author: please provide constructive suggestions for improvement (if changes are required). 

The contribution of this paper is not sufficient for publication in this journal. Authors should submit the paper to a more suitable journal or conference.

Reviewer: 3

Comments to the author: please provide constructive suggestions for improvement (if changes are required). 

The title "Low Latency Datacenter Networking" is not proper for a survey paper. It is  rather too specific for a survey article. The "low latency" of the paper is one of very old topics in communications networks. According to your article, reviewer could not find any different technologies to decrease latency compared to traditional network technologies.  Even though the topic is very specific, the survey article did not provide insightful future research topics but it mentioned only superficial several prospective views.  You categorized the topic problem solving into four technologies, reducing queue length, accelerating retransmissions, prioritizing mice flows, and exploiting multipath. But you didn't show cons and pros of four technologies. And you need to add shortest delay paths technology among physical servers or switches (routers) in datacenter. One last important thing is that you are suggested to add some experimental results for readers' better understanding.

Reviewer has found that the same survey paper under the same title had been submitted in Dec. 12, 2013 and was published in July 2014. Refer to 

There are some more comments: 

- Avoid using acronyms if possible because frequent uses of acronym interfere with readers' easy reading and understanding.

- You use too many non-general terms, for example, D^3, ECN, D^2TCP, FCT, HULL, PDQ, CP and so on.

You are suggested to use full names even though the original papers used acronyms for general readers."

My rebuttal:
To reviewer 2: I think you just provided yet another example of a totally meaningless review. I believe our paper should get another review to replace this.
To reviewer 3: An "old" topic does not justify an automatic rejection. By this standard no publication on TCP, routing, or topology should be published today. Maybe the reviewer just likes buzzwords like "fog computing" in the title to make sure the paper is novel. The comment about zero novelty of the various techniques used to reduce latency in data center networks is blatantly wrong. This comment just means that all the Sigcomm and NSDI papers surveyed in our paper do not provide any new ideas, which I couldn't believe is actually raised by a networking researcher. It seems to me the reviewer doesn't even know anything about data center networks. 
We did provide some future directions. Whether they're "superficial" or not may be arguable, but please provide reasons that make you think it's superficial. We didn't compare the four general techniques, true. Come on there's a word limit for this 6-page article in IEEE Network; we barely fit everything into 6 pages with some discussion of future directions, and you want us to even add experiment results?!
And the comment about shortest path. Yes, of course a network uses all the shortest paths available. I don't think we need this reviewer to educate us on this... Most data center network topologies have multiple equal-cost shortest paths, and that's introduced in our survey as the background.
Our draft was submitted to arXiv, yes. So what? This is a common practice. arXiv is not a journal/conference at all. (OMG maybe this reviewer doesn't know?...) We posted our article on arXiv so everybody can read our paper and provide comments, without waiting for the long review to finish.

The editor told us he tried to contact many reviewers for this article. Yet these are the reviews we receive after 8 months. Our survey can be found at my website or arXiv. We put many efforts into this survey, and I believe it is in general well-written (we didn't have the chance update it to include more recent results though because of the rejection). I think it is unfair for authors to receive irresponsible reviews like these for their work, and I believe IEEE Network has a lot to do to improve the review quality, if they care about it. 

Post has shared content
A bit more than 11 years ago we installed our 1000th rack, shown here on the loading dock.  We knew it was #1000 so we tricked it out a bit. These racks were a bit cleaner than the original corkboard racks :-)  We called them fridges because...well, I don't know why.  

Just to clarify we didn't actually have 1000 live racks at that time, since many older racks had been retired.  This was the 1000th Google-designed rack ever produced.  

Post has shared content
Two Russian rooftoppers' expeditions in Hong Kong. Beautiful pictures!

Post has shared content
The Ultimate Rube Goldberg Machine... With Dogs!

Ad Creator: "It was a very long day...."
Animated Photo

Post has shared content
1 Minute = 60 Second, definition is not the same when we say 1 internet minute. 
#Internet In One Minute :-  #Amazon. com is generating  US$83,000 in #online sales every 60 seconds. #Apple users download 48,000 apps from the App Store – Every Minute.

#Google   #Twitter   #Facebook   #Vine   #Pinterest   #Tweet   #Email   #YouTube   #Skype   #Yelp   #WhatsApp  

Post has shared content
Conference Report: USENIX Annual Technical Conference (ATC) 2013
Posted by +Murray Stokely, Google Storage Analytics Team

This year marks Google’s eleventh consecutive year as a sponsor of the USENIX Annual Technical Conference (ATC,, just one of the co-located events at USENIX Federated Conference Week (FCW,, which combines numerous conferences and workshops covering fields such as Autonomic Computing (, Feedback Computing ( and much more in an intensive week of research, trends, and community interaction.  

ATC provides a broad forum for computing systems research with an emphasis on implementations and experimental results. In addition to the Googlers presenting publications, we had two members on the program committee of ATC and several keynote speakers, invited speakers, panelists, committee members, and participants at the other co-located events at FCW. Google is proud to support the academic community through conference participation and sponsorship.

To read some highlights from ATC 2013, head over to the Research Blog, linked below.

Post has attachment
Wait while more posts are being loaded