Oct, 17 , 2012 – San Diego – The eNovance team
Day 2 at OpenStack Summit. The sun is still over here, and more news are coming. Here we go !
By the way, during all sessions we follow, we try to cover all topics in which we are interested to.
Canonical Announcement
We had this morning several interesting keynotes and I really loved the Canonical one. They made announcements about Juju product.
Juju now provides official support with OpenStack Deployment and that’s really cool. Thanks to Mark Shuttleworth, we have seen how much powerful is this software, and he successfully performed a demonstration of upgrading OpenStack from Essex to Folsom in a minute. Since this topic is a real challenge today, we need to find a way to think about continuous integration and Juju seems ready to perform that.
Another cool thing was the new dashboard of Juju management. Made 100% in HTML5, you can perform most of the Juju common tasks. If you like this spoil and want to see screenshots, you can see the pictures I took and uploaded on the eNovance Facebook page.
Nova
During the sessions, we talked about missed features:
- no per-user quotas (was reverted because bugged)
- no bare-metal provisioning but it was reverted because too big
Here is the Grizzly plan :
- Finalizing quantum switch
- It’s going too slow because we want to make sure nothing breaks
- Missing security groups in some cases
- Missing multi host
About Cells :
Some scale problems are simplified by reduce to 200/400 nodes. It already runs in production at Rackspace who is using this code. Security groups do not work with cells at the moment. There is a public branch to use it already, it should be ready early in the cycle.
Other topics mentioned : No database compute, scaling, security, upgradability, Bare metal provisioning
Better user experience for volumes :
- block device mapping is painful / cryptic
- ongoing discussions to design a new interface
- unified storage code
Other cool features :
- Configuration cleanup ( remove the configuration options that are useless )
- Policy improvement ( configuration policy.json files is clumsy and complex )
- DNS integration ( what’s the pathwork for DNS ? there is stuff in nova, there are external projects), every body wants it
- Database improvements
- More API (EC2, google compute API etc. )
Cinder
New Features that will be introduced :
- Create volume from image
- NFS file as Virtual Block Devices
What went well :
- A lot of participation : every vendor (netapp, ibm etc. ) all made new drivers for cinder
- nova-volume equivalence : nova-volume will be deprecated after this release
- core status
What’s next (this is speculative and the list is only for hot topics) :
- QoS : the ability to control IOPS because a lot of backend have this ability
- Volume State Implementation : it’s a state ( available, ready, failed, attaching etc. ). We’re going to convert into a state machine. It’s still point of contention because it’s just a string.
- API improvement : a possible version 2, driven by vendors who what to expose features
- Secure Attach
- Multiple Back-End driver support : you can only chose one backend so that the tenant can chose the backend. Tiered pricing depends on that. Will be worked on.
- Volume types : what does it mean ? Defining and collecting use cases to define what it means. The scheduler is simple minded and needs to evolve, based on types.
- Retention of meta data for glance images on bootable volumes. The loss of meta data is problematic when rebuilding etc.
- Backup to object store : cinder => swift.
- Volume resize
- Xen enhancement : from the folks on citrix.
What did not go so well ?
- Documentation is lagging behind
- We got into a mode : “it’s exactly the same as nova-volume” but we must improve
- Figuring out how the release cycle is tight. It’s a difficult to figure out how to schedule things. It’s going to be interesting how to get used to it.
- Consistency with reviewer : there is a good response time but the set of reviewer is too small (5) and we need more eyes on the code
High Availability
Florian Haas leads this session and here some points we have been talking about :
* Documation in progress (available from docs.openstack.org) but still need to me improved.
* Active / Active setup need to be documented too.
* We need to integrate the agents in OpenStack Repositories (we hope it’s coming soon).
* Need also to be integrated with Puppet, Chef and Juju to facilitate deployments.
Quantum
Most important topics we talked about were IPV6 integration for most of features in Quantum. We also discussed about Load Balancing as a Service (LBaaS). This topic has already some open-source code in many places (i.e. Mirantis), but the work need to be converged with the goal to provide a common tool to perform LBaaS in OpenStack.
Tomorrow, we will talk about RackSpace use case, but also eBay Quantum deployment and many things else…
Follow us on Twitter : @enovance
And check #OpenStack hashtag on Twitter.