10 minutes fun with Cloud API comparison
0 likes1,344 views
The document compares cloud APIs from Amazon AWS, Rackspace with OpenStack, and Google Compute Engine, focusing on the number of API calls required to visualize cloud infrastructure. AWS and Rackspace require numerous requests to retrieve information, while Google Compute Engine offers a more efficient, aggregated approach but requires OAuth authentication. The author advocates for improvements in cloud platforms to enhance performance and ease of use.
10 minutes fun with Cloud API comparison
- 1. Little fun with cloud API comparison lcerveau@nephorider.com
- 2. • Our product, Nephorider, wants to provide a one click visualization of a cloud based infrastructure. A diagram like in computer school • This include : Load Balancers, Instances (including image information), Volumes, SQL DB basic network information (e.g elastic IP) • Now how many queries should be made to get all this info?? We want the shortest time Context
- 4. Amazon AWS • No authentification phase • Principles: Action+XML • Usually requests returns all information for an category of items. The “container” contains all its content but with no details • Some questions requires additional calls. Disk image, is an IP elastic • 8 regions
- 5. Amazon AWS 8*(DescribeInstances+DescribeLoadBalancers +DescribeVolumes+DescribeDBInstances) Fixed Variables Max(NInstances*(diskimage-id)) +Max(NInstances(checkstatic)) > 42 requests, count 50
- 6. Rackspace with Openstack • Needs one pre-authentication (although it can last) • Global call may or not return all informations needed to draw the graph (cf DB) • Not always “descending logic” between “containers” and content. E.g instance to volume comes with volume info • AFAIK no concept yet of Elastic IP (see http:// feedback.rackspace.com/forums/71021-product-feedback/ suggestions/998625-configurable-ip-addresses) • 5 regions
- 7. Rackspace with Openstack Fixed • /v2/tokens + Authentification • 5 Regions (volumes/details+/servers/detail+ / instances (for DB) + /loadbalancers) Variables • NLoadBalancer *loadbalancers/LBID for instances (Node keys) • Max(NInstances*(image)) • NDB * instances/instance_id/users (for info on users) > 22 requests count like 30
- 8. Google Compute Engine • Requires an OAuth dance for authorization, then with limited token • At first optimization per region is possible through /projects/PROJECTID/regions (but apparently not zones). Returns usage per region {! "metric": "CPUS",! "limit": 24,! "usage": 1! },! • Better there are aggregated calls over all regions !!!
- 9. Google Compute Engine • APIs have fields so this allow to ask for the needed information • Inside an API call URL to the “containee” object is returned • Load balancing is not named “per se” (ForwarwindRule and TargetPool)
- 10. Google Compute Engine Fixed • /oauth2/token On Compute Engine • /project/PROJECT-ID/aggregated/instances/ • /project/PROJECT-ID/aggregated/disks/ • /project/PROJECT-ID/aggregated/targetPools • /project/PROJECT-ID/aggregated/ forwardingRules On SQL • /project/PROJECT-ID/instances/ 6 to 8 (images and IP not explored)
- 11. Conclusion • Although format changes (XML versus JSON), AWS and Rackspace are in same philosophy. Iterate over regions. • By itself Rackspace requires even more calls per region • Google Cloud Platform is more a next generation API. This really came as a “last minute surprise” • We want the other to catch up!!!!!
- 12. Thank You!