AWS gives infrastructure insights - and it is very passionate about it

• AWS sees the pace of adoption picking up  – We may see this more today as the keynotes start, but Hamilton shared the above slide – showing the expectation that the latest industry generational change is happening faster than previous one, the move of x86 servers into the cloud. And he is certainly right that the pace of this change is faster than in the past where these transitions (e.g. mainframe to UNIX servers) have taken more than a decade.

• The network is the critical path – Not surprisingly Hamilton shared that networking is the critical path for cloud, but it was interesting to see with what consequence AWS tackles the issue with a simple but compelling logic behind it: With networking being expensive, but only being less than 10% of cloud infrastructure cost, networking should not limit access and utilization of the most expensive resource in the cloud infrastructure, which are at more than 50%. Every loss of server utilization due to networking shortage costs even more as more servers need to be procured. Hamilton e.g. shared that the loss of a single IP packet equates to 0.2s loss of compute capacity…. So AWS builds its own network hardware, has its own protocol stack and runs its own private long haul links. With that the AWS team has been able to reduce network jitter significantly (and impressively).

• Massive scale – Hamilton also shared insights into a single data center (DC), Amazon runs multiple DCs per Availability Zone (AZ), and a single DC is typically over 50k servers, often up to 80k. Making a back of a napkin calculation that makes AWS a 4+ million server cloud. And that would mean that AWS runs a teenage market share number for worldwide virtualized systems (40M+) – so there is room to grow. AWS doesn’t want DCs larger because of blast radius concerns, but like them not more than a quarter mile apart.

• AWS is positively positively ‘obsessed’ – And the obsession is a good one, listen to customers and then squeeze everything out of the infrastructure and go all the extra miles to make the infrastructure better, more resilient. E.g. Hamilton shared that he has seen the very rare event of a data center outage due to a failure in switchgear equipment three times in his career, so AWS now re-writes the firmware of the switchgear in its data centers. AWS also runs custom sub power station because local utilities can’t provide fast enough reliable power infrastructure. Too many other examples to share, but this was a clear red thread through Hamilton’s presentation.

MyPOV