Enterprise rails hosting 3 ways to scale - 2011-10

Enterprise Hosting for Ruby on Rails

Julian [email protected]

http://www.enterprise-rails.de

Samstag, 15. Oktober 11

mailto:[email protected]

mailto:[email protected]



Introduction


IntroductionAbout me


About meJulian Fischer‣Twitter: http://www.twitter.com/railshoster‣E-Mail: [email protected]


http://www.twitter.com/railshoster





‣ CEO of Avarteq GmbH








‣ Lecturer „Ruby on Rails“ @ HTWdS









‣ Ruby und Ruby on Rails programmer









‣ Ruby und Ruby on Rails programmer

‣ Entperise-Rails.de - Head of Hosting






IntroductionAbout Avarteq GmbH


@railshoster

About Avarteq GmbH




@railshoster

‣ Founded in Nov. 2008 from two existing companies.

About Avarteq GmbH




@railshoster


‣ Involvment of Key-Systems GmbHmanages ~2,5 * 10^6 domains for customers of 200+ countries.

About Avarteq GmbH




@railshoster


‣ Involvment of Key-Systems GmbHmanages ~2,5 * 10^6 domains for customers of 200+ countries.

‣ Team size: ~20 people>16 full-time

About Avarteq GmbH




IntroductionAssumptions for this talk


@railshoster

Assumptions for this talk




@railshoster

‣ Ruby and Ruby on Rails architectures to be hosted.





@railshoster


‣ Hosting basics you know what a LB is and what it does.





@railshoster



‣ You are interested in a methodology to determine a suitable hosting solution for you. And you don‘t need to see Ruby code in a presentation to like it :-)





@railshoster



‣ You are interested in a methodology to determine a suitable hosting solution for you. And you don‘t need to see Ruby code in a presentation to like it :-)





Hosting todayand in the near future


The future of hostingcan be found in the cloud.


But not on EC2.


US Cloud providers do to hosting what Starbucks did to coffee.

From $1 with free refill to ...


More and more free or achievable cloud

technologies emerge.


@railshoster

Hosting today and in the future




@railshoster

‣ Eucalyptus





@railshoster

‣ Eucalyptus

‣ OpenStack





@railshoster

‣ Eucalyptus

‣ OpenStack

‣ vCloud





@railshoster

‣ Eucalyptus

‣ OpenStack

‣ vCloud

‣ ...





Many „smaller“ clouds will become available

within the next few years.


Additional abstraction layers will be established.

Helping to stay independent or even mash up existing clouds.


Abstraction via services


„Start an instance on an internal cloud, then with the same code start another on EC2 or Rackspace.

Deltacloud protects your apps from cloud API changes and incompatibilities, so you can concentrate on managing cloud

instances the way you want.“

http://incubator.apache.org/deltacloud/




Client side abstraction


Fog

„Collections provide a simplified interface, making clouds easier

to work with and switch between.“

https://github.com/geemus/fog




So be prepared for the cloud.Design your app scalable.


BUT


For nowConsider your options.

Pick carefully.


Making a decision requires options to choose from.


Know your options!


3 Hosting ScenariosOverview


@railshoster

3 Hosting Scenarios




@railshoster

‣ Cloud Hosting

3 Hosting Scenarios




@railshoster

‣ Cloud Hosting

‣ Commodity Hardware Cluster

3 Hosting Scenarios




@railshoster

‣ Cloud Hosting

‣ Commodity Hardware Cluster

‣ High End Hardware Cluster

3 Hosting Scenarios




3 Hosting ScenariosCloud Hosting


Cloud HostingUrban Legends


@railshoster

Urban Legends




@railshoster

‣ Cloud technologies are wonderful

Urban Legends




@railshoster


‣ People become euphoric

Urban Legends




@railshoster



‣ Marketing people oversimplify advantages

Urban Legends




@railshoster



‣ Marketing people oversimplify advantages

‣ People want do believe

Urban Legends




#1 Go to the cloud

and get rid of sys ops.


@railshoster

Urban Legends




@railshoster

‣ The hole decision process should be accompanied by an experienced sys op. You don‘t need plenty of workhours here but you definitely want to have advice here and there.

Urban Legends




@railshoster

‣ The hole decision process should be accompanied by an experienced sys op. You don‘t need plenty of workhours here but you definitely want to have advice here and there.

‣ You‘re still running unix services. Depending on how custom your system is you will need sys ops helping to configure your services accordingly. Template configs from the community might not fit your scenario. How much memory do you got? You have to adapt cache sizes, number of Passenger processes to be spawned, ...

Urban Legends




@railshoster

Urban Legends




@railshoster

‣ Who writes your chef recipies? Who writes your service configuration templates? Who tweaks your database?

Urban Legends




@railshoster


‣ Every got a mistery failure? Which shouldn‘t be there but actually is. Library dependency issues, sporadic network problems, spontenously failing processes, ...

Urban Legends




@railshoster


‣ Every got a mistery failure? Which shouldn‘t be there but actually is. Library dependency issues, sporadic network problems, spontenously failing processes, ...

‣ Who recovers from serious failures? Taking your backups and refills production databases.

Urban Legends




Who gets up at night?


Not having a sys op means ...


@railshoster

Urban Legends




@railshoster

‣ Making a mistakes potentially causing a lot of trouble. It‘s so hard to be a sys op and programmer at once. These roles have different perspectives, by nature.

Urban Legends




@railshoster


‣ Accpeting drawbacks by not focusing on service configurations, ...

Urban Legends




@railshoster


‣ Accpeting drawbacks by not focusing on service configurations, ...

‣ Having people acting like sys ops sbd. performing sys op tasks can be considered a sys op. Even if he/she looks like a programmer :-)

Urban Legends




#2 Go to the cloud and

have no outages any more.


@railshoster

Urban Legends




@railshoster

‣ Data center issues can‘t be avoided From experience, there will be network outages caused by switches, routes, broken wires, whatever. http://rh.gd/nMNSUK as an example.

Urban Legends




http://rh.gd/nMNSUK

http://rh.gd/nMNSUK

@railshoster


‣ Self healing can‘t recover every issue Resetting your computer often helps. Hence it‘s a good guess. But it does not help all the time. Do you really want to distract your programmers to dig through unix service logs? Ineffective.

Urban Legends




http://rh.gd/nMNSUK

http://rh.gd/nMNSUK

@railshoster


‣ Self healing can‘t recover every issue Resetting your computer often helps. Hence it‘s a good guess. But it does not help all the time. Do you really want to distract your programmers to dig through unix service logs? Ineffective.

‣ Autoscaling won‘t keep your service up unless you set it up correctly AND your autoscale settings either match the situation or be very aggressive (wasting money). Do you really want to scale out on a DDoS attack?

Urban Legends




http://rh.gd/nMNSUK

http://rh.gd/nMNSUK

#3 Go to the cloud and

save money.


@railshoster

Urban Legends




@railshoster

‣ Not true for every case. You have to do the math.

Urban Legends




@railshoster

‣ Not true for every case. You have to do the math.

‣ Actually it‘s pretty expensive.

Urban Legends




#4 Go to the cloud

it‘s safe.


@railshoster

Urban Legends




@railshoster

‣ It‘s simply not There are serious legal issues being discussed.

Urban Legends




@railshoster

‣ It‘s simply not There are serious legal issues being discussed.

‣ Infrastructure non-transparency Do you really understand the infrastructure topology? If not how can you be sure about your data being kept private?

Urban Legends




Again. Cloud is a future technology.

But don‘t be blind.


Cloud HostingPro


@railshoster

Cloud Hosting




@railshoster

‣ Maximal flexibility with Auto Scaling

Cloud Hosting




@railshoster


‣ Grow & Shrink start and stop (app/server) instances. Shut down what you don‘t need and pay per use.

Cloud Hosting




@railshoster



‣ Robustness with auto healing.

Cloud Hosting




@railshoster



‣ Robustness with auto healing.

‣ Kill and restart misbehaving components.

Cloud Hosting




Cloud HostingContra


@railshoster

Cloud Hosting




@railshoster

‣ Scale carefully or do you really want to scale out to serve requests to DDoS attack resulting into a FDoS - a financial denial of service by spending all your money? Setup your scaleout rules carefully.

Cloud Hosting




@railshoster

‣ Scale carefully or do you really want to scale out to serve requests to DDoS attack resulting into a FDoS - a financial denial of service by spending all your money? Setup your scaleout rules carefully.

‣ Autohealing within limits switch off - switch on it not always a suitable repair strategy. Nothing replaces an experienced sysop! It‘s just reducing his workload taking away no brainer tasks.

Cloud Hosting




@railshoster

Cloud Hosting




@railshoster

‣ Causes legal issues for EU companies putting data to a US providers. http://rh.gd/p6l26y (german). In certain scenarios virtualizatin is completely forbidden (e.g. certain levels of pci compliance).

Cloud Hosting




http://rh.gd/p6l26y

http://rh.gd/p6l26y

@railshoster


‣ Unreasonable expensive compared to physical hardware. 4 - 20 times more for similar performance. Depending on your cloud provider and instance/resource type.

Cloud Hosting




http://rh.gd/p6l26y

http://rh.gd/p6l26y

@railshoster


‣ Unreasonable expensive compared to physical hardware. 4 - 20 times more for similar performance. Depending on your cloud provider and instance/resource type.

‣ You are a number by being one in a million customers. Don‘t expect reasonable help when needed.

Cloud Hosting




http://rh.gd/p6l26y

http://rh.gd/p6l26y

Did you know?



3 Hosting ScenariosCommodity Hardware


Commodity HardwarePro


@railshoster

Commodity Hardware




@railshoster

‣ „Cheap“ hardware in huge datacenters.

Commodity Hardware




@railshoster


‣ Traffic included.

Commodity Hardware




@railshoster



‣ Private GBit backnet available. Physically dedicated

Commodity Hardware




@railshoster




‣ Maximum performance from your budget. Get a E3 Xenion

Quad Core 1245, 16 GB ECC RAM, 2*3TB HDD for 69 € / month.

Commodity Hardware




@railshoster

Commodity Hardware




@railshoster

‣ Why would you scale out if you already have plenty of workforce for your money? That‘s a simple thought. A very good option especially if you know your load and grow rate.

Commodity Hardware




@railshoster

‣ Why would you scale out if you already have plenty of workforce for your money? That‘s a simple thought. A very good option especially if you know your load and grow rate.

‣ No legal issues for EU companies.

Commodity Hardware




Commodity HardwareContra


@railshoster

Commodity Hardware




@railshoster

‣ You are a number by being one in a couple of thousand customers. Expect response and repair times within 30 min to 24h.

Commodity Hardware




@railshoster


‣ Scaling less flexible. Server relocation might be necessary. Server setup fees make shrink operations expensive.

Commodity Hardware




@railshoster


‣ Scaling less flexible. Server relocation might be necessary. Server setup fees make shrink operations expensive.

‣ Hardware is not high end. Smart system design needed to compensate eventual hardware failures.

Commodity Hardware




3 Hosting ScenariosHigh End Hardware


High End HardwarePro


@railshoster

High End Hardware




@railshoster

‣ High end hardware in well staffed datacenters. Close contact to data center technicians. Quick responses. Fast failure recovery.

High End Hardware




@railshoster


‣ Very realiable hardware. Hardware RAID, hot swappable hdds, redundant power supplies, ...

High End Hardware




@railshoster




High End Hardware




@railshoster




‣ Maximum performance. Get multi cpu servers with 2 x 6

XENON Cores, up to 192 GB RAM, SAS and SDD hdds, ...

High End Hardware




@railshoster

High End Hardware




@railshoster

‣ Custom equipped servers can be built specific for their individual purposes. A server with fast cpus and SSD disks will boost your DB service, for example.

High End Hardware




@railshoster


‣ Vertical scale outs Servers can be upgraded by adding more memory, better CPUs, more hdds, ...

High End Hardware




@railshoster


‣ Vertical scale outs Servers can be upgraded by adding more memory, better CPUs, more hdds, ...

‣ Hybdrid solutions with private clouds. If there‘s a private cloud co-located in the datacenter you can have physical machines to encounter ground loads and use a cloud workforce to serve peaks.

High End Hardware




@railshoster

High End Hardware




@railshoster

‣ Scale outs New servers can be added. High end data centers usually provide advanced switching technologies with vlan capabilities.

High End Hardware




@railshoster

‣ Scale outs New servers can be added. High end data centers usually provide advanced switching technologies with vlan capabilities.

‣ No legal issues for EU companies.

High End Hardware




High End HardwareContra


@railshoster

High End Hardware




@railshoster

‣ More expensive than cheap hw Still much cheaper than cloud resources for a comparable performance.

High End Hardware




@railshoster

‣ More expensive than cheap hw Still much cheaper than cloud resources for a comparable performance.

‣ Contract durations Due to financial write offs contract lifetimes are usually longer than for commodity servers or cloud resources.

High End Hardware




Proceed systematically.

Think. Evaluate. Think again.


How to build a hostingAn exemplary process


An exemplary processRequirements Analysis


@railshoster

Requirements Analysis




@railshoster

‣ Budget Calculate roughly 2.5 to 5% of your overall project costs.





@railshoster


‣ Expected Load





@railshoster


‣ Expected Load

‣ Ground load Present on a daily basis.





@railshoster


‣ Expected Load

‣ Ground load Present on a daily basis.

‣ Peak load Triggered by (external) events.





@railshoster





@railshoster

‣ Desired Response Time How many ms to serve.





@railshoster


‣ Desired Availability Uptime / Downtime ratio.





@railshoster



‣ Geographic Targeting Where are your users located?





@railshoster



‣ Geographic Targeting Where are your users located?

‣ Estimated Growth 1 month, 6 month, 12 months.





@railshoster





@railshoster

‣ Legal constraints Be aware of privacy law issues.





@railshoster


‣ Technical constraints e.g. ugly Windows legacy stuff.





@railshoster


‣ Technical constraints e.g. ugly Windows legacy stuff.

‣ Competence Know how to run an app in production?





An exemplary processIdentify System Components


@railshoster

Identify System Components




@railshoster

‣ Which components are present? search server, background processing, key/value store, replicating rdbms, ...





@railshoster

‣ Which components are present? search server, background processing, key/value store, replicating rdbms, ...

‣ Eliminate SPOFs Try to have everything crucial twice+.





An exemplary processChoose Platform


@railshoster

Choose Platform




@railshoster

‣ Which platform matches your requirements best? Shrink/Scale, cheap workforce or rock solid?

Choose Platform




@railshoster


‣ (Really) need resources on demand? Or is it just for the case that ... If you need to grow and shrink, cloud providers will to the trick.

Choose Platform




@railshoster


‣ (Really) need resources on demand? Or is it just for the case that ... If you need to grow and shrink, cloud providers will to the trick.

‣ Want to have the maximum performance from your budget? Then hardware might be your choice.

Choose Platform




@railshoster

Choose Platform




@railshoster

‣ Geographic location Where are your users.

Choose Platform




@railshoster


‣ You and the supply chain Are you a number or a customer? Do you have access to technical staff?

Choose Platform




@railshoster


‣ You and the supply chain Are you a number or a customer? Do you have access to technical staff?

‣ Quality of Service What are the response times? How fast do they repair?

Choose Platform




An exemplary processInitial Dimension


@railshoster

Initial Dimension




@railshoster

‣ Know your app To match your desired response time you‘ll need to know your app‘s resource consumption. At least roughly.

Initial Dimension




@railshoster


‣ Create a system design draft. Decide where to run which system component. The smaller your building blocks are the better.

Initial Dimension




@railshoster



‣ Estimate the # servers you need. And decide about scaleout settings if you‘re on a cloud plattform.

Initial Dimension




@railshoster



‣ Estimate the # servers you need. And decide about scaleout settings if you‘re on a cloud plattform.

‣ Order servers or whatever you rely on.

Initial Dimension




An exemplary processSystem Design


@railshoster

System Design




@railshoster

‣ Detailed System Design Describes where to run what. How to use what resource when and how. How what to connects to what.

System Design




@railshoster


‣ Monitoring plan. Monitor everything your app relies on. This will give you diagnosis information when you need it. So plan what to monitor how.

System Design




@railshoster


‣ Monitoring plan. Monitor everything your app relies on. This will give you diagnosis information when you need it. So plan what to monitor how.

‣ Backup plan. Data can be lost. Be prepared and think about where your app produces valuable data to be backuped.

System Design




An exemplary processInstallation(automation)


@railshoster

Installation(automation)




@railshoster

‣ Get a bluprint of your hosting Use Chef.




@railshoster


‣ Describe how to setup every system component. Keep your recipies small and simple.




@railshoster


‣ Describe how to setup every system component. Keep your recipies small and simple.

‣ Be ready for a deployment. You have successfully automated your installation process if you can run cap deploy afterwards.




An exemplary processStrengthen your system.


@railshoster

Strengthen your system




@railshoster

‣ Your app need dependent processes? Use a process monitor such as kickstart, monit or god to keep them alive. Double check with nagios for a system health status.





@railshoster


‣ Make your servers reboot proof. Your system needs to get back to normal on a black out automatically.





@railshoster



‣ Desaster and Recovery planning. What is the most likely error event? What is the impact? How to fix it?





@railshoster



‣ Desaster and Recovery planning. What is the most likely error event? What is the impact? How to fix it?

‣ Simulate disasters. Test your recovery plans.





An exemplary processDeployment


@railshoster





@railshoster

‣ Use a deployment tool. Not really worth mentioning but just to be complete. Do I have to mention capistrano?





@railshoster

‣ Use a deployment tool. Not really worth mentioning but just to be complete. Do I have to mention capistrano?

‣ Setup Capistrano. Cap deploy. Shouldn‘t need more than that for the average case. Automate whatever you can. Make it foolproof.





An exemplary processTest && Launch


@railshoster

Test && Launch




@railshoster

‣ Test your setup. Perform acceptance tests. Partly covered by

your desaster simulations. Also test with your customer, your users, ...

Test && Launch




@railshoster

‣ Test your setup. Perform acceptance tests. Partly covered by

your desaster simulations. Also test with your customer, your users, ...

‣ Go live! Cover your eyes and pray :-)

Test && Launch




Lessons LearnedKey Statements to remember


@railshoster

Lessons Learned




@railshoster

‣ Creating a hosting. Should be considered to be a project. Process it systematically.

Lessons Learned


@railshoster


‣ Be aware of people holding a hammer. Because if you got a hammer everything you see is a nail.

Lessons Learned


@railshoster



‣ Make decisions. No decision without choices. Know your options.

Lessons Learned


@railshoster



‣ Make decisions. No decision without choices. Know your options.

‣ Reduce sysop work. But running an app is a pain and burdon for most developers.

Lessons Learned


Focus on adding value to your business.


Focus on your app.



Thank youfor your

attention!

Headquarter:http://www.avarteq.de

Blog:http://ww.treibstofff.de

Rails Enterprise Hosting:http://www.enterprise-

rails.com

Rails Hosting:http://www.railshoster.com


http://www.avarteq.de




http://ww.treibstofff.de

http://ww.treibstofff.de









Technology

Enterprise rails hosting 3 ways to scale - 2011-10