[cs615asa] Meetup Summary

[Kareem Mohamed]

I attended the nycdevops meetup on April 17th.

https://www.meetup.com/nycdevops/events/fmgjmnyxgbwb/

I mostly picked this because it featured a talk on "serverless"
architecture and I curious as to exactly what this entailed and 
what
benefits it provided.

The meet featured two talks, "Breaking down Dev/Prod Wall" and 
"Crossing
the Serverless Fireswamp".

The first talk given by Andrew Phillips focused on the 
relationship
between Dev and Devops and talked about how it was essentially the 
same
issue we see with Dev and Management. The speaker proposed that we 
stop
just hurling tasks over "the wall" at each other and instead try 
to
cross-pollinate with each other. He basically reinforced that 
building
reliable software is inherently a social activity and making sure 
that
you interact with all stages of the pipeline is crucial. A nice 
point I
found he made is that you should try to actually get to know at 
least
one person on the Management/Dev/Devops side, try to learn more 
about
the problems they think about everyday and what your side of 
things can
be doing to make their job easier.

The second talk given by Mike Roberts spoke about "Serverless
Architecture" and some of the pitfalls associated with it. 
Essentially
Serverless is outsourcing all of your back-end infrastructure or
frequently run functions to cloud providers like Amazon and 
Google. The
idea is that they are much more equipped to handle it because of 
their
size and experience with scalability. The speaker mentions AWS 
Lambda as
a prime example of a service that can be scaled to massive 
proportions,
being able to have AWS automatically fire up thousands of 
containers and
fire off your function in parallel in response to spikes in 
traffic is
really nice. He notes however that you must be careful because 
those
thousand containers are now all running functions that are likely
hitting your DB, which is probably not equipped to handle that 
much
traffic at once. He calls this issue "unmatched scaling" and urges 
us to
make sure that we try to use AWS Lambda's limiting features to try 
to
mitigate it from over-scaling and hammering your core-endpoints.

He also spoke about the benefits of being able to distribute your
infrastructure across the entire world. It's extremely unlikely 
they
AWS's us-east region goes down at the same time as the us-west 
region
or the eu regions since they operate separately. Thus by putting 
your
infrastructure across multiple regions helps mitigate any 
potential
downtime seen with using the Serverless model.

The main takeaways I got from this talk are that Serverless is 
really
good at scaling fast and being cost effective. It however is not 
good
for time critical issues or services that require a server to be
"always-on" since the AWS Lambda model requires a "cold-start" 
time for
every container spawned. However, for non-time critical tasks AWS 
Lambda
seems like a good tool to look into.

He mentioned he's in the process of rewriting his old article on 
the
topic available at 
https://martinfowler.com/articles/serverless.html.