2009/3/23 van Sleeuwen, Berry <berry.vansleeu...@atosorigin.com>: > A reason for having DASD not set to shared is to make use of MDC (our > selling point is that we do not want guests to cache filesystem data > because we have MDC). But when guests are running in multiple VM systems > would this hurt in any way? I do know that whenever a guest has to be
I think you reverse the argument. You can't tell Linux not to cache data because it is a fundamental part of how things work. You can configure a Linux server to reduce the resources it spends to cache data, but that's no guarantee that it will not cache data. Whether you have MDC on or off does not change the way Linux works with its own page cache. You could mount a file system with an option to disable page cache, but performance implications of that are normally very unpleasant. When your Linux virtual servers share certain disks (a common r/o file system) then you must not write to them. If you do, both Linux page cache and MDC may give you out-of-date data and cause data corruption. Just don't write to them. Use some approach where you prepare a new disk and put it in place through directory updates. Even when Linux disks are private for the virtual server (so not shared with other virtual servers) the a small amount of MDC can be helpful in some cases. Eg to compensate for the fact that Linux tracks don't match hardware tracks and when Linux made the wrong decision not to cache certain important things. I'm not sure you would actually be able to measure it. Unless your infrastructure guarantees that Linux will not find old data in MDC, I would recommend to stay away from it. Defining the Linux volumes as "shared" is an easy trick to disable MDC. If you look at the value of data cache in general, the first layer of cache picks the low hanging fruit. So the I/O you get after Linux page cache are already less favorable from a cache point of view. You would need an excessive amount of memory devoted to MDC to make a difference there. Considering the fact that most I/O subsystems already have such an excessive amount of memory anyway, throwing in a GB of MDC may not help you a lot. This also applies to a common R/O shared Linux disk. -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
