Matty wrote: > On Sat, 13 Aug 2005, Tony Nguyen wrote: > >> Ryan, >> >> Sorry for a late reply but things are quite busy around here. Sounds >> like you're really digging into the SVM/UFS area :^) >> >> Now, to the question you asked. You're quite right that four 8K I/Os >> should be on the be same disks if we use stripe width of 32K. >> However, this is the best case scenario where all I/O to the stripe is >> of size 8K. The author, on the other hand, looks at the case where >> the I/Os are put on the disks with an offset (first I/O is not aligned >> on the stripe boundary). Thus, four 8K I/Os would always involve >> another disk. This assumption is not always correct but is probably >> closer to most environment where I/Os are of different sizes. The >> stripe width/interlace size we'd recommend is 512k and the resync >> buffer size of 512k. These values work quite well for most systems. > > > Will the default values in Solaris be adjusted to these anytime soon? >
It'll be in Nevada but not S10 since we can't introduce this type of change (interface change) in a micro release. Should I notify you when the nevada bits containing the change is available? >> >> Let us know if you have additional questions. >> -tony >> >> Matty wrote: >> >>> >>> Does anyone happen to know what the author means by "stripe width has >>> a probability of splitting 1 of 4 I/O" and "whereas a 1-Mbyte stripe >>> width splits only 1 of 128 I/O operations": >>> >>> http://www.informit.com/articles/article.asp?p=169475&seqNum=2 >>> >>> "The probability of splitting an I/O operation is inversely >>> proportional to the stripe width. Consider an OLTP system which >>> mostly performs 8-Kbyte I/O. A 32-Kbyte stripe width has a >>> probability of splitting 1 of 4 I/O operations, whereas a 1-Mbyte >>> stripe width splits only 1 of 128 I/O operations (less than one >>> percent). Increasing the stripe width is the single most important >>> improvement you can make to decrease the probability of splitting an >>> I/O operation." >>> >>> If four 8k I/Os are issued, I would have thought that they would have >>> been written to a single spindle (based on an interlaced value of >>> 32k). Not real sure what he means by "splitting" operations? >>> >>> Thanks for any insight, >>> - Ryan >>> _______________________________________________ >>> lvm-discuss mailing list >>> lvm-discuss at opensolaris.org >> >>
