Cheryl Watson writes: >And I strongly believe that a parallel sysplex environment >that leaves the applications available to users 100% of the >time is the best solution. Once you implement such a >configuration, it doesn't require you to keep any given >LPAR up for several months at a time.
To expand on Cheryl's excellent point, there is an advantage to Parallel Sysplex (including single frame implementations) in terms of software maintainability. The preventive maintenance policy that Cheryl describes is a very reasonable one, but some single production LPAR customers still find it difficult to follow because they are understandably reluctant to schedule a user service interruption. Parallel Sysplex offers distinct advantages in supporting reasonable preventive maintenance practices, rolling version and release upgrades, and so on. It provides a very useful "safety net." Of course it's still important to test any sort of software changes before rolling them into production. And, of course, the benefits of Parallel Sysplex (single frame or otherwise) must be weighed against the (comparatively modest) costs. I think Parallel Sysplex is presently under-exploited (on average, in general). But things are changing. I should also mention that, even short of a Parallel Sysplex, it is possible for a single frame customer to have at least one cold or warm standby LPAR in addition to the "live" production LPAR. It depends on the application architecture(s) how useful this is, but it is always at least somewhat useful if you cannot get to a Parallel Sysplex implementation immediately. I'm not sure what to call this approach -- maybe a "TogglePlex"? Assuming reasonable operations skill you can reduce service outage windows substantially with a TogglePlex approach. Basically you apply whatever software updates you want to the standby LPAR, declare a ~2 minute "soft outage" at 2:00 a.m. Sunday morning (or whenever), and, at that time, point all users over to the updated LPAR. That now becomes the production LPAR. The next time there's a service update, toggle in the other direction. With DVIPA, automation, and other dynamic features the outage can be quite minimal or even imperceptible, depending on the application and middleware. And, if necessary, you can fall back much faster, because you keep the backlevel LPAR up and running (at the old software level) until you're convinced everything is fine with the new level LPAR. With a TogglePlex you don't force users to wait for applying maintenance or IPLs. In fact, you can stretch out the IPL within a softcap if helpful. (There's no rush.) And any IPL problems will not cause user interruption, because you're not taking any IPLs during the window. So you can do as many IPL rehearsals as you want, too. Anybody using a TogglePlex approach today? (And what do you call it? Is there a better name?) What sort of outage window is the result in your particular environment? Apologies if this suggestion is just way too obvious, but perhaps there are some newer folks unfamiliar with this (probably old-school) approach. - - - - - Timothy Sipples IBM Consulting Enterprise Software Architect Based in Tokyo, Serving IBM Japan / Asia-Pacific E-Mail: [email protected] ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
