HN Theater @HNTheaterMonth

James Hamilton [0][1] and others [2][3][4] have given pretty good overview over the years of the AWS infrastructure.

[0] AWS Innovation at Scale (2014), https://youtu.be/JIQETrFC_SQ

[1] Tuesday Night Live (2016), https://youtu.be/AyOAjFNPAbA

[2] Operation of the AWS Global Network (2019), https://youtu.be/UObQZ3R9_4c

[3] Foundations of AWS Infrastructure (2018), https://youtu.be/gH46jrFfiCc

[4] Exploring the AWS Global Network (2018), https://youtu.be/tPUl96EEFps

⬐ WoahNoun

He's also written about AWS infrastructure (and electricity infrastructure in general) on his blog. Fascinating stuff.

https://perspectives.mvdirona.com/2017/04/how-many-data-cent...

https://perspectives.mvdirona.com/2017/12/when-you-cant-affo...

https://perspectives.mvdirona.com/2017/04/at-scale-rare-even...

https://perspectives.mvdirona.com/2018/11/1227/

https://perspectives.mvdirona.com/2019/02/aws-nitro-system/

http://m.youtube.com/watch?v=JIQETrFC_SQ

I think this is the right talk by him, but there's one where he implies some differences. For instance, that because they can promise exact temperature ranges they can clock them differently. My guess is same fab, just binned differently, or maybe different packaging.

⬐ ComodoHacker

Perhaps with overclocking unlocked via microcode.

James Hamilton gives a good talk that goes over networking (and more). This is it from reinvent 2014.

https://m.youtube.com/watch?v=JIQETrFC_SQ

Around about here in the talk: https://youtu.be/JIQETrFC_SQ?t=14m38s James Hamilton talks precisely what the definitions of AZs and regions are :)

This looks great. They also have the advantage that James Hamilton mentioned AMZ got from building their own hardware - the get to do a cleansheet s/w design and not have to support all the gazillion options that Cisco does.

The talk was on HN a little while ago, worth watching: https://www.youtube.com/watch?v=JIQETrFC_SQ

The scale at FB means they're probably saving hundreds of millions doing this.

> This suggests to me that running distributed databases in multiple AZ's has almost no latency penalty (e.g. you won't even paying eventual consistency / replication lag taxes that you might think are a danger).

Yes, Hamilton points that out in his talk: https://www.youtube.com/watch?v=JIQETrFC_SQ

It has been, or at least the talk that seems to be what this article is on.

https://www.youtube.com/watch?v=JIQETrFC_SQ

⬐ patman81

Perfect, just what I was looking for!

⬐ pwarner

These guys are the new x86. On that note I was surprised they actually get custom chips from Intel, but I guess it makes sense. At the AWS summit in SF earlier this year the only AWS supplier with a booth was Intel.

⬐ mad44

In summary, AWS rides the benefits of economies of scale. (http://en.wikipedia.org/wiki/Economies_of_scale)

They design/build their networking gear, full hw/sw stack. This is cheaper and more reliable (their code is simple/customized to their datacenter use case.)

They also have SingleRoot I/O virtualization at each server: each guest VM gets its own hardware virtualized link, which is great for reducing the giant tail at scale problem (google for Jeff Dean's description of the problem.)

Their relational DB system RDS is getting popular: 40% of customers using them. So they compete with Oracle by offering similar highly-available service with much less price. They keep adding new relational DBs: Aurora, RedShift, EBS.

They design/build their power infrastructure. Faster.

They are very customer oriented, they make things simple/painless for customer use cases. They are obsessed with metrics, measuring everything, with tight feedback loops to improve things weekly. They rolled 449 new services + major features in 2014 alone.

⬐ mbesto

Actually, all of what you just mentioned is considered "economies of scope" (http://en.wikipedia.org/wiki/Economies_of_scope)

Here, economies of scope make product diversification efficient if they are based on the common and recurrent use of proprietary know-how or on an indivisible physical asset.