This morning I’m thinking through what I’m going to say at the AWS re:Invent Conference next week and I suppose I’m in a distractable mood. When an email came in asking “what advice would you give someone entering the information technology industry in 2017?” I gave it some thought. This was my short take:
Play for the long term. Choose jobs working on technology that will still be relevant a decade hence. Choose jobs that build on your strengths but significantly stretch beyond them. Always say “yes” to requests to do more or take on more. Each is an opportunity. Choose jobs working with the best in the industry. Working with the best is the quickest way to learn. Never let go of the details even as you take on broader roles. Don’t worry about money, job title, or recognition. It’ll all come and never leave if you get these four goals right. Short term decisions often yield little and what they do offer doesn’t last. Play the long game.
If you are comming to re:Invent, I hope to see you at my talk at 8pm on Tuesday (11/29) night (Tuesday Night Live with James Hamilton). If you aren’t going to be in Vegas, it’ll also be live streamed.
Happy holidays and I hope to see you next week at AWS re:Invent.
I watched the live stream and really enjoyed it (I went back and forth on whether to attend re:Invent this year, and really regret choosing not to). One thing that struck me, though, was your segment on the drive towards powering AWS with renewables.
You made a comment at one point along the lines of “if people would just use US West Oregon, we could get there more quickly”. I’d LOVE to move my stuff to Oregon, but as an Alexa developer I’m handcuffed by the Alexa -> Lambda functionality only working in Virginia. I’m sure there are financial reasons to restrict Alexa to only a few datacenters, but does the renewables consideration come in to play in deciding where to deploy that functionality?
Thanks Eric. Which Alexa functionality that you need is currently only in Virginia?
Hi James,
Do you have any piece (already written) on Quantum Computing? If not, may I know your thoughts on the same?
Quantum Computing has the potential to fundamentally change what can be done computationally. What’s really hard to predict is when. There are many quantum systems in operation but none of these are able to run at a scale that supports solving problems not practically solvable by conventional von Neumann architectures. Many of these early quantum systems are interesting and they help establish the effectiveness of different approaches but we are not yet at the point where the systems are producing results we can’t get using conventional systems.
When that happens, it’s is going to be huge. Potentially the biggest change ever in the computing world.
“distractable mood”; yes the tower of silence that is entering the whitehouse in 2017 is distractingly abritrary.
Ahem, Your admonishment on to ruminate upon long term (read a broad set of skills) employment and not to prioritze ornamental titles and wages is spot-on. I’d like to thank you for it. A cursory look at the AWS re:Invent Conference agenda, leads to the obvious question of cyber security and how if ever all those SoCs(smartphone, tablets, IoT) could ever be actually “secured” in the broadest sense. But perhaps that is a fiscal matter for the manufactures of IoT. Also cryptocurrencies (https://en.wikipedia.org/wiki/List_of_cryptocurrencies) what is Amazon’s stance and why; there are more that 700 of such.
thanks I’m very glad the event will be live streamed.
Mark, one recommendation I should have added to my short list was curiosity. I’m interested in absolutely all forms of technology and will grab every change I can to learn the details from another field or to see how other types of systems are manufactured. It’s amazing how frequently I learn something that can be applied to problems we are working on in infrastructure. The more solutions one sees, the more likely some of these ideas are to fall together in a new way to solve a problem in infrastructure. I’m interested in all forms technology and many of the best ideas lie in new applications of technology from other disciplines.
Your concern on IoT security is a good one given the massive DDoS from a couple of weeks back. To be sure the industry will see more of these but, as hard as the problem looks, I’m confident that the industry will make progress. Technically nothing in security is ever “solved” — we just make it harder and less rewarding to do bad things. But I’m very confident that will happen and IoT devices will get hardened fast. It’s too bad that we are going to see more of these IoT hosted events before we see solution but I’m confident it will happen.
Your related question on crypto currencies is a good one. We are interested in crypto currencies and, even more interested, in the application of the base technology of blockchains and systems of distributed trust that can be built upon blockchains.
Returning to the specialized application of blockchains in cryptocurrencies, some interesting trends emerge. All of the early cryptocurrency mining operations ran on general purpose processors and many ran on AWS. The high scale operations have since moved to using custom ASICs. What makes this interesting at least to me is I think this move to hardware is a trend we are going to see repeated in most feilds.
Everyone has been saying that software is going to eat the world. For sure, it is true. Software will be at the core of everything. But the most successful systems are going to be stradling software and hardware. Software will be everywhere but the hotest and most frequently executed kernels of these software sysetms will be implemented in specialized logic. FPGAs and specialised ASICs are going to be increasingly important.
Hi James I look forward to this talk as i am sure it will be great like the prior one 2 years ago. It would be fantastic if in your talk you could address what you think are unique attributes to AWS infrastructure, including perhaps some cost and PUE comparisons, how do you see AWS infrastructure 2-3 years from now, including how AWS infra will cater to emerging workloads like deep learning, an outlook for cloud infrastructure given difficulties in Moore’s Law in some aspects of the infra like CPUs and optics. I also hope you make your slides available in this website. Thanks a lot
Eric, your questions are in line with my interests. Looking at PUE, we have improved annually every year since I’ve been there and likely every year before that as well. That’s not especially surprising but it does show that enough big changes are being made in the AWS infrastructure every year to keep pushing the number down. Most of the industry best are currently in the 1.12 range (+/- a point or two). You might see one outlier reporting out at 1.08 but PUE is an imperfect measure and it can be gamed. I ignore the 1.08 claim and just keep focused on findig more we can do to improve efficiency.
Over the years, many of the big efficiency gains have come from the mechanical side mostly because that where it has been easiest to make improvements. We work on both mechanical and electrical but mechanical had the biggest opportunity so that is where we found the biggest gains. As the cooling overheads get down as low as they are now, my focus has swung to electrical distribution and I expect most of our gains over the next 2 years will be there.
You were asking about optics as well. The rise of merchant silicon routing ASIC has put router costs on a Moore’s law curve. Optics have most deffinitely not been on a Moore’s law curve so, over time, networking costs has been swinging to optics. On the optics front, the big news is silicon photonics. Lasers, detectors, and logic in silicon on the same die. The “on the same die” part is still comming soon but there are nice multi-chip module solutions available today and same die solutions appear near. This will put network optics on a Moore’s law cost reduction curve as well.
Silicon photonics are hugely important in what they will allow in networking but, even bigger and more important, we are one step nearer to optics directly to CPU. This will allow many core system innovations amoung them the ability to virtualize memory (without latency penalty). Other possibilities opened up by silicon photonics are arguably bigger innovations but virtualized memory alone would be huge.
Optics direct to the CPU package has always been “10 years out” but it’s feeling much closer to reality today. Perhaps only 10 years out :-).