- MySQL Cluster Data Warehouse Edition. HP converted Tandem NonStep into HP NewView. MySQL Cluster can do the same. I am sure this will happen, but probably not by next year.
- Running a cluster of database servers on 1 big SMP. Sun has the hardware and software to make this possible. The new Sun T2+ server has 32 cores and 256 hardware threads. Solaris has lots of features for partitioning a larger server. The use of SMP as a cluster has been written about before. This is a good fit for MySQL as it will be a while before MySQL can effectively use a large SMP and multi-threaded MySQL Cluster will be limited to small (less than 10) number of threads.
- MySQL performance on Intel SSD. Actually, this should be open to any SSD vendor that provides commodity pricing and great random-write performance. This platform significantly raises the bar for what we expect. Sun has written about using Flash as a cache between memory and spinning disk. It will be interesting to see how this turns out with respect to price performance.
- How we pushed changes from Drizzle back to mainline. Dare to dream. For now I will settle for the changes to compile with -Wall and remove compiler warnings.
- High-throughput reporting performance: MySQL, ZFS, Flash and Sun Thumper. Thumpers have been popular for running Greenplum Postgres, but have not garnered as much press for MySQL. I distinguish reporting from data warehouse workloads. In this case there will be many concurrent complex queries. The queries have enough selectivity that nested loops join is preferred over hash join. 48 disks in a box should give you 5000 to 10,000 IOPs.
Saturday, October 18, 2008
MySQL Conference proposals that I want to see
These are proposals that I would love to see for the 2009 MySQL Conference:
Subscribe to:
Post Comments (Atom)


re MySQL Cluster Data Warehouse Edition
ReplyDeletemissing features
1) distributed join processing
we have started working on this (not only thinking) but (currently) focused on making small joins faster. (and I don't refer to BKA, but about actually doing parallel join-execution)
2) index on disk-data
we started on refactoring needed to add this
3) various compression techniques
not started, but we're discussing it
4) other index types (i.e bitmap indexes)
not started
anything else that we lack?