Meet the Devs returned this October with a trip to University of Birmingham (OK, so I didn't have to travel very far...).
This time, we focussed the day as a workshop on cloud and Spectrum Scale, Dean Hildebrand (IBM Research, Master Inventor, Senior Research Scientist) was in the UK speaking at another event and we were lucky enough to catch him for a meet the devs. We were also joined by John Lewars (IBM, STSM) who flew in pretty much for the day, and also MTD usuals, Ulf Troppens and Dan Kidger.
For those not familiar with Dean, he worked on pNFS, AFM and the OpenStack integration when they were all part of IBM Research. I've been lucky enough to know Dean for a couple of years and he was extremely helpful with getting the right contacts when I was developing CLIMB (www.climb.ac.uk) a few years back and using Spectrum Scale as an infrastructure.
The morning kicked off with Dean presenting an overview of Scale with cloud, for example the different use cases; Scale as an infrastructure service, Scale access from inside VMs and containers.
Dean gave a great talk on containerisation and covered some ideas they are kicking around to improve container support with Scale, for example looking at how you can move containers from local storage to shared and then be able to benefit from mmclone (as is used with the OpenStack Cinder drivers). We also had a discussion on data security into the container and how this might be improved going forward.
John picked up the batten with a fascinating discussion on FPO and shared nothing environments. It was really interesting to hear how FPO, which was originally designed for hadoop workloads is being used as a PoC with storage rich servers in the public cloud. John went over different placement methods and stressed that its only performant for very specific read-intensive workloads as well as covering data security with storage that can't be twin tailed (as in the public cloud).
As lunch was running late, I gave a quick overview of some of the challenges we have getting Spectrum Scale data INTO a VM (as opposed to running on it), and we talked about the lack of Manilla support for multi-tenant networks. One idea we did kick around was running Ganesha in namespaces with different configs, it was suggested we might approach this by using containers.
Pizza arrived a little late (as usual!), and thanks go to OCF for supplying a more than ample array of pizza, and perhaps caused a localised potato shortage from the wedges ...
The afternoon turned to discussion of AFM, we covered a number of use cases and the AFM technology used to support them. I'm always amazed at the number of different ways people are using it to support their workloads, and I know its inspired one of my team to look at AFM between sites to solve a specific problem.
Finally John wrapped up the day with some slides on his work with running Spectrum Scale in Amazon cloud and their ideas about trying to get an AMI into the marketplace to support this for limited deployments/testing. Again the focus was on the lack of support for twin tailed storage, and we had some fun debate on placement of EBS volumes (for example if we replicate with Scale, how do we know Amazon aren't still placing that on the same physical drive under the EBS volume, so is the data really safer?).
Thanks again to IBM for supporting the meet the devs and getting developers out to speak to our user communities, and also to OCF for buying the pizza.
Finally, I'm closing with a rant. Registration - if you register to come to an event, please take it seriously! Cancelling the day before doesn't really give us time to offer the place to someone else! So please please, if you register, turn up!
We have plans in place for a MTD in 2017, so watch out on the mailing list for event details.
I'll leave you with a comment Dean made, "we give you the nuts and bolts, and you take them and build them into services", which I think embodies my experience of working with Spectrum Scale.