arrow-down coffee engineering consultancy development remote-management support linkedin twitter youtube email phone gitlab github

AWS re:invent 2017 recap

Submitted by walterheck on December 04, 2017

As I’m sitting here at McCarran Airport in Las Vegas my brain is going crazy. I just finished attending AWS Re:invent and just like last year, it was an exhausting, frustrating, exhilarating, breathtaking, mind widening experience. Here’s some notes from this week.

The Good

Lots of things make this conference amazing. A list of things that come to mind:

Chalk talks

Chalk talks got a much more prominent place this year, and they are awesome. How often do you get the chance to ask an AWS network expert on how exactly advanced VPC design should be done? These sessions were often guided by a handful of slides to set the topic and then were open to any and all questions.

Deep learning summit

I signed up for this on a whim, not knowing much of what to expect. This turned out to be one of the most interesting sessions: 4 hours of talks by some of the world’s smartest AI/Deep Learning/Robotics experts from places like CalTech and MIT explaining in-depth what they are working on, which problems they are facing and what is around the corner. The last session in this summit was done by a manager of a VC fund that specializes in these fields. Special praise for his self-intro: “I’m the dumbest person on this stage, but I have lots of money to make up for it”. The good thing about this last session was that it brought all the theory (much of which was far too advanced to grasp in detail) back down to a real level: he talked about what kind of ventures they are funding and why, and what kind of problems they would like to see solved. 

New announcements

Obviously re:invent wouldn’t be re:invent if it didn’t have tons of new services introduced. The specific tech that is interesting is different for everyone and there’s tons of info out there on the web, so I’ll keep this short to my personal favorites:

  • cross region VPC peering: previously VPC peering only worked inside a region and crossing regions required all kinds of complexity. This is now very simple.
  • Go in Lambda: I started my career as a Delphi programmer and Go has won my heart by providing a lot of the features Delphi had: single binary compilation, strong type system, decent error handling. Needless to say I’m happy to see Go added as a supported language for AWS Lambda
  • VPC Endpoints: As you may or may not realize, some AWS services do or live inside a VPC, like for instance S3. That meant previously that if you wanted to reach those services from a private subnet you had to find a way to reach these services that involved going outside the VPC. With VPC Endpoints this is no longer needed.
  • Private Link
  • Transcribe, Translate, Comprehend: this set of services was introduced during the keynote by Andy Jassy. They are individual services that allow understanding of spoken text, transcribing it to text. Then translating that text with impressive accuracy and lastly comprehend which is able to extract intent from text. Imagine integrating this with the rest of AWS’s services and a whole new world opens up.
  • Recognition video: AWS already offered rekognition for images which worked surprisingly well. Now they did the same for video. A demo on stage was downright impressive
  • Neptunus: I have a strong love for databases and specifically for RDBMS style databases. Many things can be logically modeled in relational databases, but graphs are not one of those things. Neptunus is a graph database as a service, opening up a whole host of interesting options (asking your database for a shortest path between A and B which returns that path and all the nodes in between for instance)
  • Aurora Multi-master: Again, stemming from my database love (big fan of MySQL) I was happy to hear that they somehow pulled off Aurora multi-master. I’ll be curious to dig in deeper and see how it holds up under stress and/or error scenarios
  • EKS/Fargate: As many people already expected, AWS announced their managed Kubernetes offering named EKS, putting ECS in a strange spot. Besides just EKS they also announced Fargate, an attempt in removing the need to worry about clusters and machines for Kubernetes. That should reduce the need for skilled Kubernetes engineering, even though initial reviews question the price point

The food

This one goes in both categories. Some of the food was really good, but things like breakfast were..well.. up for improvement. Special mention for the military style precision with which food was organized in the Venetian food hall. A hall that could probably hold 20.000 people where no matter how many people walked in at the same time, you always have food without queueing. Very impressive.

The AWS Benelux party

Since last year some of the companies from the benelux region (Belgium, Netherlands and Luxemburg) decided to throw a party just open for people from the region. This is a very nice place to talk about AWS (and many other things of course) with people from the same geographic area.

The bad & Ugly

Unfortunately a lot of things were very frustrating as well, some of them causing me to lose almost a full day of schedule on Wednesday. A list:

The mobile app

From uncashed schedules to long loading times to schedules appearing empty out of the blue, on to an annoying full screen ad every time you activate the app. Worst for me was that in order to even try to sign up for a session you had to sign out of any other sessions you had scheduled for that slot. If it then turned out your new session was full, you had just successfully screwed yourself out of both sessions. For a conference of this size and AWS being the company they are they need to do a lot better next year.

The full sessions for non-fanboys/girls

Unless you bought your tickets on the day they became available and signed up immediately for the sessions you wanted, there was no way to get into them. I have a busy life and customers to deal with, so many of the sessions I wanted to get into were full already. They tried to counter that by reserving 25% of seats for each session for walk ups, but for the popular sessions people easily got in line 3 hours before their session. 

The queues

Further to the point above, there were lines for absolutely everything, leading to much frustration. I’ll just leave it at that.

The spread out venues

The conference was spread out across many venues along the Las Vegas strip.  This made it so that you could’ve just run back and forward between sessions, but you had to choose sessions based on the venue they were in. Even changing venues once a day meant 30 minutes best case scenario and up to one and a half hours worst case scenario spent on transiting. Again, I expect better.

Global Partner Summit

This year was the first year the global partner summit was open for everyone, not just partners. This was a bad choice in my opinion. As a partner we work hard all year for our partnership (and we’re very happy to do that) but by removing the exclusivity from these sessions the dynamic changes, and it also sends a message with regards to appreciation for the work we do.

In addition, the GPS welcome party was a failure: thousands of people (most of whom were not partners) in a line meant that I wasn’t able to enter the venue until 7:15 pm (with a scheduled time for the whole reception from 6pm-8pm). When I finally got inside, the venue turned out to be a night club with the music volume turned up to 11. I didn’t come to party, I came to talk to other partners and have meaningful conversations.

Favorite sessions

A lot of good content was provided. Some of my personal favorite sessions:

Advanced VPC sessions

A lot of the consulting and engineering services we provide for AWS require a deep understanding of one of the most complicated and advanced parts of AWS: the networking that underpins it all. I saw several sessions on advanced VPC topics that really helped me understand things better.

Deep Learning Summit

I covered this earlier in the blog post, but I want to touch quickly on some of the topics discussed. One of the speakers showed research they have done where they fed a ton of images along with spoken descriptions of them into a machine learning environment. Without ever teaching the algorithm how to recognize words it managed to make surprisingly accurate guesses which part of the audio discussed which part of the image (eg. Everytime the sound for the word face comes by, there’s a face in the picture).

Another speaker showed their journey to get a robot to complete the pacific crest trail (leading from the border of mexico along the west coast to the border of Canada) faster then the fastest human had done so far. This provides interesting challenges.

A third speaker showed how they trained a machine learning algorithm using images and recordings of the accompanying sounds. After a while the algorithm was able to point out which parts of the image caused which sounds (eg. everytime you hear water flowing, there’s water in the picture).

Og-aws 

The open guide to AWS is a community-driven effort to curate AWS documentation. The current docs are already impressive and the community behind this is really friendly. More help is always welcome, you can join their slack here: https://og-aws-slack.lexikon.io/ 

Midnight madness talk with Andy Jessy

This was a very informal session with AWS CEO Andy Jassy on the Sunday night. It was very interesting for the entrepreneur side of me to listen to someone in his position explain how AWS got started (1 EC2 availability zone in one region with one OS and one instance size) and give some tips on what they do to be so successful (they write a press release and and FAQ at the very start of a new project to make sure everything in the project serves the right mission).

Mechanical Turk loading of data into AI

One of the services I never paid much attention to is Mechanical Turk. This session showed how to use Mechanical Turk to collect training data for a Machine Learning algorithm to be trained. Provided you set up he units of work correctly you can get lots of training data very fast. One of the best tips: remember the people executing the tasks are human beings and treat them fairly. The Mechanical Turk community will blacklist your work quickly if you treat them unfairly.

Conclusion

Just like last year, the conference was a great experience overall. I learned a lot, met a lot of new people, discussed a lot of AWS things and just in generally had a good time. I would be happy if I could avoid some of the frustration but overall for a conference this size it was worth the time and investment.

Now, I’m happy this was my last travel for the year, time to spend some time with the family!