Ask HN: What big companies use PaaS services and why?

A full time devops person is one person defining and managing the processes that will put your software on servers and keep them running and scaling. A PaaS provides you with dozens or hundreds of such people along with stable general-purpose platforms and processes that already work.<p>PaaS starts at pennies per hour and employees start at thousands per month on top of which you still need the servers etc they will manage. Cost can be prohibitive both ways.

The company I work for has deployed an on-premise PaaS using CloudFoundry. You get many of the benefits of a PaaS without the astronomical costs.<p>Of course, we do need a small team of excellent DevOps engineers to keep the whole thing moving forward.<p>They are also now starting to deploy Openstack in order to benefit from a software-defined infrastructure. It&#x27;s all very cool.

I would love to use a public PaaS, but cost is a bit, shall we say, steep ATM though.<p>We run around 550 instances of apps, totalling at ~250Gb RAM. So ~450Mb RAM per instance. In the coming moths that will just grow, as more and more of the online applications gets either migrated or written for running in the PaaS.<p>With Heroku[1], the pricing example stops at 150 dynos, but if the pricing is linear we would spend $19657&#x2F;month.<p>With Bluemix[2] $14295&#x2F;month.<p>Im sure that with that amount of instances one could negotiate better terms.<p>Ofc, hypervisors and storage is something you have to factor in the equation. But being a large enterprise we already have heaps of that stuff lying around and being under-utilised.<p>Also, been a while(~ 1 year) since I did a performance comparison[3] between in house and public, and last time I did, our performance crushed the public offerings, so we would have to run more instances to get the same performance, bringing up the cost.<p>Another consideration we would have to take into account is the network, our applications are written with ms latency in mind, so to move into a public offering we would have to migrate all our DBs and services around the apps into a public provider near the public offering, which would be insanely expensive, complicated and time consuming. And ofc, rewrite all the applications to live in a world where we dont have &lt;1ms connectivity to everything.<p>I would say that I currently spend a couple of hours&#x2F;month actually working on CF fixing bugs, trouble shooting etc. Rest of the time I can work on other stuff, so I would not really factor in my salary to much into the equation.<p>We get compute and flash storage for &quot;free&quot; from other projects(massive SAP and other back office installations wont go away anytime soon). We get a predictable awesome network across two DCs with &lt;1ms latency for free (massive SAP and other back office installations wont go away anytime soon). We have full control over a free, open source, general purpose PaaS(CF wont go away anytime soon). Im not spending enough time administrating it(really, CF is great in that respect) to factor into the cost calculations. This VS &gt;10k a month is why we don&#x27;t use a public provider.<p>I suspect that the situation is similar in a lot of large companies, there is resources, already there for other big projects not being used.<p>[1] <a href="https:&#x2F;&#x2F;www.heroku.com&#x2F;pricing" rel="nofollow">https:&#x2F;&#x2F;www.heroku.com&#x2F;pricing</a><p>[2] <a href="https:&#x2F;&#x2F;console.ng.bluemix.net&#x2F;pricing&#x2F;" rel="nofollow">https:&#x2F;&#x2F;console.ng.bluemix.net&#x2F;pricing&#x2F;</a><p>[3] Highly unscientific, running toy apps and JMeter against them. Comparing 5 instances in inhouse CF vs 5 dynos for instance, most R&#x2F;S wins.