[cs615asa] Meetup Summary

Neel Patel npate27 at stevens.edu
Sun Apr 22 16:11:13 EDT 2018


*Meetup Information:Date: April 17, 2018 Topics: Breaking down Dev/Prod
Wall + Crossing the Serverless FireswampLink:
https://www.meetup.com/nycdevops/events/fmgjmnyxgbwb/
<https://www.meetup.com/nycdevops/events/fmgjmnyxgbwb/> I recently attended
the meetup in NYC organized by the nycdevops Meetup group. One of the
co-organizers, Tom Limoncelli, wrote one of the textbooks for our class,
"The Practice of System and Network Administration," which is one of the
reasons this particular group caught my attention. While I was interested
in learning about both talks, the main reason for attending this particular
talk was for the second topic: "Crossing the Serverless Fireswamp". The
first reason was that I did not know much aboutserverless architectures
before this meetup; I was hoping to gain more knowledgeon their pros and
cons. Another reason was because of how the topic connectedmore to systems
administration. Talk 1: Breaking down Dev/Prod WallThis topic was presented
by Andrew Phillips. In this brief talk, Phillips describes the relation
between development, operations and product management. Often in the
industry, different teams/departments are separated by a "wall," which
represents how teams may have differentreporting hierarchies and lack of
interest in the work/understanding of others, which would impact the
product's outcome. The two walls he describedwere:- The Dev/Ops wall
between development and operations, which can lead to a poorly performing
product. While it is still an issue today, it's been improving thanks to
Devops, the philosophies, culture and interests of these two areaswere
combined.- The Prod/Dev wall between product management and devops. This
wall was themain focus of the talk since is a more recent problem compared
to the Dev/Ops wall. This can result in having a product with no
customer/user base, which Phillips states is worse than having a badly
performing product. Solutions he offers include communication, visibility
and transparency between the developers the PMs. Also, involving the user
the ideation processcan help the quality of the product and interaction
between teams. While all of this may seem like simple solutions on paper,
it is still can difficult to pull properly in reality. I learned that the
issues that plague common product/software development canalso be applied
to systems administration, since PMs and the systems administrators can try
to learn more about each other's interests and incentives in order to align
them, so that the entire team/department/companycan improve their product.
 Talk 2: Crossing the Serverless FirewswampThis topic was presented by Mike
Roberts. It was a more technical oriented talk and had a bunch of
references to "The Princess Bride" movie in it. Roberts described two types
of serverless servers:-Backend as a Service (BaaS)-Functions as a Service
(FaaS) The topic focused on serverless FaaS using AWS Lambda as a main
example. AWS Lambda is a service relatively new service (4 years old) that
allows oneto load instances of functions that are event-driven. Just like
S3, there arebenefits to using Lambda, which Roberts talks about:-
Scalability  - Multi-region support and availability- No need by the
developer to provision servers- Cost. The cost is based on usage, so the
functions are set up but no traffic is going through, you will not get
charged. Also there are less labor costs involved for the companies using
them. It is cheap to test the waters out too and see if it is a good fit
for the company or not.- Speed. Rapid conversion from ideation to a working
product can be done within days. Also it's a great for rapid prototyping.
Roberts personally likes to use them it for data processing as it allows
fora flexible data pipeline. He also goes over some of the downsides of
using serverless FaaS with Lambda and how to tackle them:- Unmatched
scaling. In a multi-layer architecture, this can be an issue, since
different layers may scale differently. For example, having numerous Lambda
functions updating the same single SQL database can cause some issues. A
potential solution in this example would be to use a similarly scaling
database like DyanamoDb, limit the number of instances of a function by
setting concurrency limits. or use a message queue.  - Cold starts. When
running functions for the first time, instantiation may take some time. If
it is a negligible amount of time or a very rare occurrence, it is not too
big of a concern. However by using the right tools, it is possible to
reduce cold starts. An example he gave was that running a JVM is probably
not going to be as fast as using python. Overall, I learned about what
Serverless FaaS means, what problems to expectin the future when using this
architecture, and how to handle them. It was interesting to hear about how
additional services like AWS Lambda at Edge use AWS CloudFront CDN to push
more of an application to the client, making it not only serverless, but
also regionless. He also gave some situations whereserverless might be a
better option to companies, such as when a company is small and more
innovative. Roberts wrote an in-depth piece on serverlessarchitectures that
expands on what was said in the talk. The link is below, definitely an
interesting read if you want to look more into the
topic:https://martinfowler.com/articles/serverless.html
<https://martinfowler.com/articles/serverless.html> *
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.stevens.edu/pipermail/cs615asa/attachments/20180422/83b82eb3/attachment-0001.html>


More information about the cs615asa mailing list