Answer by Ted Dunning · Sep 27, 2011 at 06:31 AM
It is quite doable but we really don't recommend it.
First, if you do this, you have two levels of management of the disks.
Second, it obscures the view of the hardware so that the auto-diagnostics don't work right.
Third, it is just weird.
Can you say a bit more about why you want to do it this way? Why not just take the disks away from LVM and use them directly?
Answer by yufeldman · Sep 27, 2011 at 02:02 PM
Take a look at: http://mapr.com/doc/display/MapR/Setting+Up+Disks+for+MapR under Section "Working with a Logical Volume Manager" if you still decide to use LVM
Answer by Ted Dunning · Oct 03, 2011 at 07:39 AM
This should be a separate question. I will repeat the answer there if you will put it in as a separate question.
The answer is that the performance of MapR mounted via NFS on localhost should be awesome. Mounted on a remote machine, the performance should only be limited by your networking. We use it internally for home directories and to support mySQL and it works very well.
For really large data (larger than you should be storing in mongo or mysql), you may be able to discern lower error rates on the locally mounted NFS server because it will do checksumming before transiting the network. That will give you robust end-to-end detection of data corruption.
Answer by sirpy · Oct 03, 2011 at 12:53 AM
I need to store files from other sources on the same partition, which are not related to mapr/hadoop. how would you describe the performance of mapr with direct nfs for daily usage?as a local filesystem for the computer holding that partition? the partition will be used mainly for storage of large files. also would it have good performance as a filesystem holding a mysql/mongodb database?
Support for >2TB disks on Centos 5.x 1 Answer