Merge pull request #19 from zalando/feature/sizing-advice

Wrote some chapters for sizing information and scaling up/down.

docs/admin-guide/scaling.md

@@ -0,0 +1,19 @@
+# Scaling up or down
+
+At some point during the lifetime of Spilo you may decide that you may need to scale up, scale down or change other
+parts of the infrastructure.
+
+The following plan will accomplish this:
+
+1. Change the CloudFormation template to reflect the changes in infrastructure
+2. Increase the number of EC2 instances in your AutoScalingGroup (or: Terminate one of the streaming replica's)
+3. Kill one of the replica's running on the old configuration
+4. Repeat step 2 and 3 until all replica's are running the new configuration
+5. Fail over the Spilo cluster
+6. Terminate the replica (the previous master)
+
+Only during the fail over will there be some downtime for the master instance.
+
+If your databases' size is significant, these steps may take some time to complete. Optimizations of this process
+(think of using EBS snapshots) are on the roadmap but not yet available.
+

docs/admin-guide/sizing.md

@@ -0,0 +1,47 @@
+# Sizing your Spilo
+
+As every use case of a database is unique we cannot simply advise you to use a certain instance type for Spilo.
+To get some notion of what is available, look at the Amazon resources:
+
+* [Amazon EC2 Instances](https://aws.amazon.com/ec2/instance-types/)
+
+# Guidelines
+We can however provide some guidelines in choosing your instance types.
+Choosing an instances type has some basic parameters:
+
+- database size
+- database usage pattern
+- performance requirements
+- usage characteristics
+- costs
+
+## Costs
+To have some notion of the costs of Spilo, we have calculated the costs for some possible configurations. 
+
+**Note**: These are costs for an individual instance, as a case study. All prerequisites (Senza, Etcd) are considered 
+to be available and are not part of the costs of Spilo.
+These figures are meant to show the approximate costs, they *will* vary depending on your usecase and resource usage.
+
+The following table shows ballpark figures of the raw costs of a single node of Spilo, as well as the Spilo costs when
+using the default configuration of 3 members in the AutoScalingGroup.
+The pricing information is taken from Amazon, and are for On-Demand pricing (August 2015):
+
+* [Amazon EC2 Pricing](https://aws.amazon.com/ec2/pricing/)
+* [Amazon EBS Pricing](https://aws.amazon.com/ebs/pricing/)
+
+| EC2   |   Storage                     |  Monthly ($) | Spilo ($) |
+|----------|---------------------------:|---------:|-------:
+| t2.micro |  10 GB EBS                 |   11     |    33 |
+| m4.large | 100 GB EBS                 |  111     |   333 |
+| r3.2xlarge | (provisioned) 500 GB EBS |  664     |  1992 |
+| d2.8xlarge | (RAID 10) 24 TB HDD      | 4292     | 12876 |
+
+## Database usage
+For some types of databases the smallest possible instance available can be a good candidate. Some examples that
+can probably run fine on t2.micro instances are:
+
+* Very small (< 500 MB) databases 
+* Databases which are not queried often or heavily (some configuration store, an application registry)
+* Proof of Concepts / Prototyping
+
+

mkdocs.yml

@@ -10,3 +10,6 @@ pages:
     - Deploy spilo: user-guide/deploy_spilo.md
     - Deploy etcd: user-guide/deploy_etcd.md
     - Connect to spilo: user-guide/connect_spilo.md
+  - Administrator Guide:
+    - Sizing spilo: admin-guide/sizing.md
+    - Scaling up or down: admin-guide/scaling.md