always create the locations that need frequent I/O /home, swap on the outer tracks ?
in [Setup]
|
From: Doug Freyburger on 19 Jul 2010 11:37 annalissa wrote: > > The following is what i have read in a magazine named "linux for > you" , To what extent this is true ? > > The general rule is that you should always create the locations that > need frequent I/O -/home, swap on the outer tracks , the easiest way > to achieve this is to create these partitions first when partitioning > your hard disk ? > > Reason all modern H.D.D's use a concept called ZCAV(zonal constant > angular velocity). this takes advantage of the fact that more linear > space is available on the outer tracks of the disk platter rather than > on the inside tracks. now since the disk spins at a constant rate , > which is also known as CAV (constant angular velocity) the read/write > I/O speed will be greater at the outer tracks as compared to the inner > tracks I tune large Oracle database and I've studied the numbers on this. The gain is a tiny percentage. If you need tuning at this level you already put each mount point on its own LUN/device and you don't use partitioning on your high access data. The gain from separating to a new device is far larger than shuffling block locations inside a single disk. Going to striped logical volumes also beats any gain from block placement. So does using an array with internal cache. While the advice is true the amount gained from block placement is tiny compared to many other tuning tactics. By the time it's worth the effort of block placement you will have already run the profiler on your custom apps to get graphs of how many time each line of code was executed. If you don't have a performance graph on your apps it is a complete waste of your time to work on tuning that will have an effect two orders of magnitude down from the top performance hogs on your system. If you have done every other type of tuning and you still need to do block placement you've spent so much time on tuning labor you could have afforded dozens of replacement servers. Use io statistics to count traffic per mount point, then place the blocks based on the actual statistics on your own host. There's no way /home will see more traffic than the basic mount points. I bet most of my systems see the traffic more like /usr, /tmp, /var, /, /opt, /home with swap and /boot appearing dead last. For the boot devices that is - Most of my systems see more traffic to the database/app devices than to the boot device. Standard tuning methodology - Find out what the bottleneck is now and work on that. If you think block placement is your current bottleneck then you have not studied the performance of your system enough to know what is really happening. Taking wild guesses and doing stuff that sounds good is far less effective than gathering the actual performance data and working from the data. Linux performance rocks when I review the data!
|
Pages: 1 Prev: A newbie question - trying to resize a given 15GB vdi file to abigger one Next: swap tuning |