[talk] Go and C++ Performance Improvements
Sujit K M
kmsujit at gmail.com
Thu Nov 2 03:06:18 EDT 2017
Firstly Thanks For such a good pointer or 's.
On Thu, Nov 2, 2017 at 3:35 AM, Edward Capriolo <edlinuxguru at gmail.com> wrote:
> Hadoop/Spark and similar tools typically divide long running processes
> minutes/hours into set of tasks. There failure model is typically that a
> failure of task only sets the process back small amounts. For example if I
> am filtering a 12GB file and I divide it into 12x 1GB sized tasks, the
> failure of task 10 can be retried without forcing the other tasks to retry.
I am more interested in knowing about the timelines in Milli Seconds. I have
worked on an Back Up Product Which is based on similar model on the Web
Side which Had about 100 MS requirements. Though I must say If a User A
starts a process of back up and server 10 is checked for then it remains during
the entire process with server 10, also there were requirements that when a
user is created it is allocated a server, you would find atleast
1000's of servers and
on operation needs we generally add 3 - 4 servers every day., It could
last hours if
the back up is huge, but the thing was it was done on the web. Though
It might take
hours It still sends across the updates on the Back up Process to the
user, these were
in millisecond time lines.
> For the three projects you are mentioning two of them look like web based
> micro-service architectures. These are typically for short processes that
> have small state and no failover.
It could still be having fail over if we use a docker container for
example or isn't
> I think in those cases you are looking for something closer to erlang's
> actor model.
> Java- Akka
> C++- c-actor framework https://actor-framework.org/
Yes this is some thing I would look for.
> In c-actor framework you can create services on different nodes and those
> services are transparently connected. Data passed between the services and
> if components die they can be re-spawned and then the framework will wire
> the service back together.
> Note: c-actor framework is not a web-server, or http based micro-service
Are there any other Low Level Latency Models like Actor Model illustrated. Also
could you please let me know whether Model Means a Design Pattern or Multiple
More information about the talk