Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Autotune – Cloud Diagnostics and Performance Tuning (acksin.com)
28 points by abhiyerra on Aug 15, 2016 | hide | past | favorite | 20 comments


> However, we found no tool to help tune servers so we decided to write one.

I actually looked for a tool similar to this a while ago after finding some larger than necessary servers in our infra not doing very much. Here are some of them:

Cloudyn Optimizer https://www.cloudyn.com/products/

CloudCheckr http://cloudcheckr.com/spend-optimization/

BotMetric Cloud Cost Management https://www.botmetric.com/aws-cloud-cost-management-cost-ana...

Most of these are aimed at cost reduction through many means (e.g, switch cloud providers, purchase reserved instances), not just improved server efficiency.


Any tool that reduced the environmental footprint of servers is a tool I want people to use. :)


Naming your software after another high-profile piece of commercial software seems inadvisable.


Good point... I actually thought the other Autotune was a piece of hardware...


It can be in a rack-mount for live use but in offhand conversation, it's an audio software / plug-in.

For future reference, the United States Patent and Trademark Office offers a lookup service to quickly check if a planned mark might already be registered.

http://www.uspto.gov/trademarks-application-process/search-t...


Given that "Autotune the news" renamed itself to "Songify the news", it's highly likely that Autotune's owners are not shy about enforcing their trademark.


I figured that move was more about wanting to promote their own brand instead of someone else's. (They began selling Songify apps around the same time.)


I first thought the article was about using the cloud to improve Autotune (the singing tool) settings.


It would be nice if a tool could dynamically tune based on server conditions, too much sequential io? Increase read ahead and swap out applications, too much random io, reduce read ahead, etc.


That is the eventual goal of this tool. But we have to start somewhere right?


No idea how this will help my idle server! If it's idle, tuning won't really help heh


Well the goal of the tool is to watch the entire cluster and notify you if any machine is idling or not doing its part.

Right now the tool is focused on making existing servers performant, but since it is already collecting diagnostic information it can eventually also notify if a machine is sitting idle.

For example, a company I worked at had a Redis slave machine which was quite powerful but we realized 2 years later that it wasn't actually doing anything. Whoops. All the monitoring metrics showed that it was performing well, but nothing actually notified us that by performing well it wasn't doing anything in the first place.

This seems to be a common theme when many machines are run. Some go unnoticed and idle away.


Excellent idea! Are there any benchmarks of how the changed sysctl values affect performance?


I've used some of the settings at a previous company on an haproxy front end machine. Lead us to move from c4.8xlarge machines to m4.xlarge handling the same traffic. I'll do a future post on how it affects scaling.


My instance isn't reporting anything. Also compains the install.sh is wrong...


Hey can you email me at abhi@acksin.com and I'll help you debug.


When is the server portion likely to be open sourced?


Hopefully in the next 2 weeks. We are considering just being able to do it via AWS Lambda.


Really frustrating when websites just plaster me in the face with an email signup form before I even have a chance to read what the business / app is all about: http://i.nick.sg/1e71733605744a9ba084737429cc694d.png

I just closed the tab. It's not worth it.


Sorry about that. Removed the form.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: