2008 Should be fun
orbist 060000HPM5 Comment (1) Visits (4556)
Just back from a nice relaxing break over the holidays, while keeping one eye on things since the start of the week. It was nice to start back on a Friday though, as it meant I could catch up on the email, sort out a few hosts and drives that had died under the onslaught from two solid weeks of performance benchmarking!
Well, the Windows host was actually killed by my own good hands last night, while trying to uninstall some stubborn software that wouldn't un-install due to running services. A quick thrash through \Cur
Only one drive failure was a nice surprise however. My ageing DS4300/4400 controllers were replaced during the latter half of 2007 with DS4700 (half as many as before for the same performance) Normally after a two week extended run doing sequential throughput matrices, single vdisk resonse curves, full saturation response curves and extended four corner tests the old (4 year old) storage was starting to struggle after its hard life in performance testing. While there are only about 120 less physical disks in my new configuration, the usual fallout rate was more like 4 or 5. I guess I'm still in the early life failure period for these drives, maybe come the summer holidays the stand will survive the same tests un-scathed.
Anyway, after some analysis of the results of the latest internal build shows some interesting areas to investigate further in some new code that has been recently introduced to the build. Thats the next few weeks of work sorted for the performance team :)
2008 really does look like its going to be interesting. As has been mentioned in various marketing releases by the SVC marketing team, it will see the long awaiting Thin Provisioning functions in SVC - although internally we refer to this as "Space Efficient Volumes". Due to the way the software stack is built in SVC, this means we get Space Efficient FlashCopy for virtually free at the same time. There are a few other key functions that are being delivered during the next 12 months and plenty for the development and test teams to be more than fully occupied with items out further in the roadmap. A few side projects (as there always are) and of course our new friends in Israel.
Well at least I managed five paragraphs before mentioning it. I do find it kind of amusing that EMC for example seem to announce a new acquisition what seems like every other week, and they get a few mentions here and there in this blogging-circle, IBM makes a storage acquisition and its all over the place. Must have really shaken up a few thoughts and minds wondering what we are up to, what plans we have for it and where its heading. Well, the press release covers what our plans for it are in the near term - longer than that I obviously couldn't comment. I have been discussing with various people the architecture and possible future applications. It certainly turns the generally applied storage controller idea on its head, in more ways than just one, and at first glance the commonality between the SVC appliance model and Nextra are quite close. Both using a flavour of Linux and Intel platform, but both doing things in very different ways.
While BarryB uses this to get yet another series of FUD comments in about SVC - I don't really see where the scalability issues you mention come from (other than [vendor removed - you can probably guess which one though] driven customer proof of concepts that look to have been deliberately misconfigured to show products in a bad light), and the only person I've ever hear mention this lack of end to end integrity is you. In all the customer visits, technical briefings, deep dive discussions, its never come up once. Maybe being such a "sym-dude" the 'mainframe' world has muddied your thoughts. However, as a developer at heart and having been in the storage industry long enough, I too agree with your comments regarding data loss. The two phrases that we dare not ever mention, and are classed as blasphemy are "data-loss" and "miscompare". Even 1 bit being lost in the middle of a file is one bit too far. Thats where we've come from and thats where we will always be. With SVC we maybe have an easier job, as we don't directly get the blame when a disk fails, or an array is lost - thats the controller thats had a problem, but we do endless - and I mean endless, as I type many hundreds of thousands of I/O streams are whirring away on our testfloors, running the latest code above and beyond the call of duty, byte checking every single word. Should something not come back as it should do, the stand is locked down, and not released until the culprit has been found. More often than not, we discover problems elsewhere - OS's, HBA drivers, Controllers (especially under stressful error inject)
Anyway, I've come to expect the usual stories and tales of woe from Mr Burke, and its good to see nothing's changed in 2008. As I said, 2008 should be fun :)