ORIGINALLY POSTED 21st Decvember 2007

10,182 views on developerworks

The house is feeling quite Christmas-e, with the tree and lights having been up for a few weeks, the collection of presents under the tree growing and the usual ‘spring cleaning’ sessions before the big day. One more day to go, and yet another working year draws to a close. I thought I’d take a few moments to reflect on what we’d achieved this year with SVC and the labs.

The first real visible accomplishment was the 4.2.0 code release back in June, which added support for the next generation 2145-8G4 node hardware using two Dual-Core Xeon processors and much faster bus and memory speeds. In addition, multiple target FlashCopy, RAS Email functions, Command audit logging and enhanced user role management. Support for 19 new storage controller models, 4 new operating systems and too many HBA models to mention!

The real work for this began late in 2006, with the new hardware meaning we could double the number of I/O threads in each node and bind a single thread to each FC-port. The development team worked for several months fine tuning the code paths, batching and binding and I spent most of my time benchmarking these changes and giving a thumbs up or down to various experimental modifications. The Cache destage algorithms were tweaked to provide what I call a ‘feedback loop’ that monitors how well the storage is coping with the work SVC is asking of it, ramping up or down the destage rates accordingly. The end result was worth it, with all of my internal benchmark four corner cases and mixed workloads showing a 100% increase over the previous 8F4 hardware. These code enhancements were not restricted to the new 8G4 hardware. In some cases over 50% increase in throughput was measured on the older 8F4 nodes, and between 10% and 20% on the original 4F2 and 8F2 nodes. The net result of this was the 74% increase (over the previous published SPC-1 benchmark) as recorded by the Storage Performance Council SPC-1 benchmark and over 50% in SPC-2 – here an 8 node cluster sustaining over 10GB/s in one of the sub-tests – the first such recorded milestone by the SPC.

The second release this year was just last month with the 4.2.1 code stream. This further enhanced the Copy Services suites by adding Incremental and Cascaded FlashCopy, user configurable CopyServices limits and double the Cluster addressability. Support for AIX 6.1 and 3 new storage controller models. My time was spent working with the Cache development team, designing, coding and fine tuning the Cache Partitioning code. This radically changed the destage algorithms again and provides an unbelievable improvement when things go bad on the back-end, meanwhile maintaining the same performance and throughput characteristics of the previous 4.2.0 code release. 10,000 nodes shipped and 3,400 clusters in daily production.

Meanwhile our lab space was expanding and I’ve been managing the upgrades. With our main lab space out of power and cooling (there simply was no more power available without running a completely new supply from the substation to this building) we grew our footprint at the other side of the site. Last weekend we had a power down and the new 400A distribution board and Ethernet subnets were installed, the culmination of over 12 months planning finally seeing the light of day. If the expansion continues at the same rate as this year this new supply should see us good for at least… erm well… 3 months?! Hehe.

Anyway, all that remains is to wish all of you a great time over the Holiday season, enjoy the time with the family and have a great New Year. Thanks for taking the time to stop by and read my ramblings, and I’ll be back with more in the New year, with Space Efficient Vdisks and Flash Copy and some other goddies coming, 2008 should be just a good.

Barry