Presenter: Scott Carlson, PayPal
This post is Sponsored by – My Linked Profile
I’ve been really excited about this session because of all of the fuss caused by the miss-announced release that Paypal was abandoning VMware for OpenStack. Scott is a Sr. Engineer at Paypal responsible for rolling out their internal Cloud, which runs OpenStack on ESX.
They choose vSphere 5.0, VCE VBLOCK because if fit nicely with their application architecture. They initially intended to use VMware’s API’s to provide the customization for their needs. During this time Paypal needed to decide on their Cloud strategy and went through another round of requirements.
The primary driver was the ability to become agile in their response for demand. They needed to run at “Cloud” spends. One of their main challenges is the unpredictability of their ownership by eBay. When eBay runs a special that includes a “Paypal Buy Now” option their traffic can double without warning. So, they are the classic elastic use case for the Cloud computing model. They therefore, had a goal of a provisioning request for a specific application to be fulfilled in 15 minutes from start to customer facing.
They wanted to create the perception for their internal customers (developers) that they had infinite capacity. PayPal felt that OpenStack provided them the flexibility to meet their goals.
When the decision to go with OpenStack, it took 60 days to have 1 pool taking traffic by the holiday season. It was a very raw implementation without a true Cloud portal. A couple of early lessons happened early on. They started to question where they used the types of compute where in their infrastructure. Paypal’s frontend is highly dense. They had over 3000
VBLOCKS VM’s running on VBLOCKs that provided 5 9’s availability. What they realized is that this wasn’t a good use case for vSphere because there wasn’t a need for DRS, vMotion or HA in this environment. However, their backend needed vSphere capability. This brought the question how to merge the capability and provide a simple portal for their internal customers.
To solve the compute challenge they leveraged the modular capability of OpenStack to manage multiple hypervisors. From a technical perspective they use multiple NOVA controllers to control each hypervisor environment. For more technical detail on installing and integrating NOVA and vSphere Ken Hui already has an excellent series on the topic. One consideration worth mentioning for this multi-hypervisor use case is that you are limited to where images can run. You couldn’t use native OpenStack to deploy images from one hypervisor to another.
PayPal plans to have 20% of their traffic running through OpenStack with both ESXi and KVM underneath for the upcoming holiday season. They are early adopters of this model so there isn’t a lot of history our other’s to look for support in their efforts. They have on slate to do at scale testing for this solution. One goal for example is to have an OpenStack certified VBLOCK.
For today they use the OpenStack Horizon portal for self-service. They plan on leveraging Netflix’s Asgard.
Paypal is definitely on the cutting edge of not just vSphere on OpenStack but OpenStack in general.