xen dev blog

Going to miss our target launch date for xen I think. I've been working non stop on it even over Christmas. Matt has been doing a fair bit of work as well.

Database remains the major stumbling block. We use master-slave replication at the moment which has served us quite well. Xenforo doesn't natively support master-slave. There are addons but they are outdated and no longer maintained. Writing our own remains a possibility although master-slave is not a particularly good model to begin with.

I've been working with mysql cluster on the new servers to try provide a theoretically more stable platform that xen will run on, however having spent four straight days working with it now I'm thinking it's too fiddly for us. Essentially it is designed as a distributed memory database which is not appropriate for the size of our database. Even if I commissioned enough ram to hold our current database ( > 20Gb of ram ) it wouldn't allow for strong growth. It does have a disk storage option which I am testing at the moment but from what I've seen so far it is more cancer.

I haven't yet decided what to do. Mysql cluster was my first choice, and we may still go with it if I can figure out how to get disk storage to work cleanly and quickly. Failing that I think I will try mariadb cluster. Failing that we might try some sort of multi-master non-cluster setup.

Long story short: I'm working on it as quickly as I can but it's taking longer than expected. In the meantime I will continue to make critical patches to the existing production system. I am still planning to move us as soon as humanly possible.

 

Different applications. We need high performance and redundancy together. It's not sufficient to have one really big server and run a single database instance on it. It would bottleneck no matter how expensive the hardware and doesn't provide a good failover option. Granted you could have two that check eachother but that's still not a great solution.

We need a scaling system made up of lots of reasonably small servers that work together -- shards. A system where we can add more capacity as needed without touching existing production hardware. The second you have to open hardware up you're talking serious major downtime and expense. If you can keep everything redundant with no reliance on any specific machine or piece hardware that's going to be the best overall option for uptime.

 

I've burned mysql cluster. Looking at mariadb cluster today

 

MariaDb cluster looks a lot more promising so far