This morning I set off from Manchester to London to run my first 'Meet the Devs' GPFS User Group session concerned about two things: Firstly, my wife is due for our second child in 5 weeks - let's hope she doesn't go into labour whilst I'm a few hours away. Secondly, I felt the fear of being an imposter, accompanied by the adrenaline rush I get when I go on stage and perform. Am I really in a position to have a fruitful meeting with a set of very experienced and capable GPFS users? Am I prepared enough? Am I a GPFS imposter?
Think. Prepare. Rehearse. And breathe.
I suddenly felt a lot better. But, I'm a manager these days and as a result less hands on, I've never used GPFS as a real user/administrator and my background is C++ development not in system administrator or architecting solutions.
I write this blog on the train home after typing up six sides of A4 scribbled notes from our first 'meet the devs' session. I chaired our prototype 'meet the devs' GPFS user group meeting in IBM Southbank's client centre.
I had the pleasure of meeting real users first hand and demonstrating where we are taking the product. I really enjoyed the sessions and graduated from being a development manager to a community member with a deeper connection with how people use GPFS and the ways they wish it would improve.
Jez (the community chair) and I set 3 hours and a planned agenda for the event. Passionate people have a lot to say (which is great) so we went off topic many times and overshot the planned time by 30 minutes. Great. Aside from a very tasty lunch and growing our networks, what key things did we learn from the day? What was the value? We all agreed the next session should be optionally longer.
I shall let the clients share their own feelings on value. From my perspective I learned more about client needs in 3.5 hours than I have done in the past six months as a development manager. I'm not ashamed to admit this fact, because I believe everyone can benefit from getting out there and talking to like-minded individuals, users and having engaging discussion. It is then we can listen for need and then collaboratively build improvements into our products and services.
I can't talk in detail about what was discussed in this forum, but I can share some key themes and insights.
Firstly, we have a product with some great attributes (ILM, fault tolerance, consistency of performance etc) and our focus on making GPFS more consumable was well received.
This group of users don't believe that making initial installations and configurations simpler is that important - the reason is that contractors and skilled people are always required for such business critical decisions. However, there was a broad agreement that making things easier to configure and more diagnosis and system information readily available is spot on. Make data available and making day to day administration consumable should be the priority.
The pricing model is an area users were particularly vocal on and would love to see more flexibility - the general feeling is that it's too binary. Why do I need to pay full whack for a sleeping node e.g. Quorum node?
If users could have a magic wand then we would suddenly have self-destructing files (without having to write your own policies & scripts etc), de-duplication, Mac OS/OSX support and in-flight encryption amongst other things.
I felt particularly 'agile' today, as we got real stakeholder feedback on install and performance monitoring work that is 'work in progress' for a future release. Generally, it appears we are moving in the right direction and there was broad agreement that the serious administrator wants the ability to access more data if needed and any simple tooling that helps automatically collate that etc is welcome. It would be nice to easily tell if someone is 'ls'ing a directory with a million files and query stats across multiple clients to get stats for a job as a whole.
Any users who take part in our demos of unreleased work can make a difference to the direction the product takes. I cannot guarantee any suggestions made will be done, but with awareness we are at least presented with the choice to take action.
Other quick thoughts from users that I can share ... don't forget SELinux is used - don't make life difficult for those people, 'who is doing what' queries would be really helpful, ensure future install CLI type commands can be called like an API and don't change across versions, it would be lovely if GUIs showed data flow and not just the raw information, can we use up to 50k filesets please, how well does GPFS scale to 25k nodes and so on.
Am I a GPFS imposter? No more than last year where I rehearsed for hours for the part of Mickey Johnston in Blood Brothers. I started the play as a 7 year old - at first it felt really uncomfortable but with exposure and practise it all came good in the end. The only way I'll feel I'm a true GPFS (IBM Spectrum Scale) community user and expert is by continuing to get stuck in, listen, learn and get hands on with it.
Big thank you to Jez, Patrick, Robin, Paul, Mark, Robert, Jess and Keith for being honest with your thoughts and making it a truly memorable day for me. We shall certainly be running more user events around the country - we can even manage pizza rather than a fancy buffet if people want to join me in Manchester city centre.
Did my wife go into labour? Thankfully not and I'm very happy to wait another 5 weeks before I hear the feedback from my new baby.
Please note that all my views are personal and I make no commitment on behalf of any organisations. What I can commit to personally is running more events and spreading the feedback I collated.