Large Filesystem I/O performence

Latest response

Hi,

I am going to make fresh installation RHEL 7.1 with ext4 filesystem. I will use data disk on storage (EMC) I want to know which better for I/O performence: many disks total 10TB (example 20 numbers 500GB disks) or one disk (10TB)

Note1: In future this system will require more capacitiy.
Note2: SAP with Oracle will run on this system.
Note3: Backup solution is Symantec Netbackup.

Responses

first point is that,, RHEL 7.x is designed to deliver best performance with default xfs file system, so why are u planning to go for ext4? ... performance wise with large file systems xfs would be a perfect fit..

I will migrate old system to new hardware. Old system use ext4 filesystem. I think no problem it to be xfs. the main issue i can not decide how configure LVM.

I agree with Murthy Sadashiva, XFS is a better choice. Did you need assistance setting up the LVM?

20 *500GB is better in term of setup. 1st , assuming your FS full in future. You just request another 1 * 500GB to extend the FS. All your LUN size will be 500GB each 2nd, multiple LUN will have better performance in term of disk saturation 3rd, for Oracle DB. You have many data files. Assume datafile01 stored in LUN001, datafile100 stored in LUN200. When user do query ( usually is read ), the READ operation will read different datafiles. This is how you saturate the Disk IO and get the max performance/throughput.

If you have 1 * 10TB,
when the FS full, you want to request 1 * 10TB ? or 1TB ? this will make the setup messy. You have the different LUN with different size.

In the strictest of terms, one needn't request a new LUN at all. A quality storage array - say, anything claiming to be "enterprise" class made in the last 15 years - will allow the extending of a LUN. Both EXTn and XFS support growing the filesystem onto the newly-available space within the existing LUN.

Fun thing with "best" questions: you really need to know the work-load you're supporting.

For example: you've seen people on this thread stating "use XFS, it's better than EXTn". Weeeellll... That really depends on what you're looking to do. If you do some googling around, you'll find use-cases where XFS is actually non-trivially less-performant than EXTn is.

Since your specific use-case is Oracle, one should probably ask the question of why you're not looking to do ASM - at least for the disks hosting Oracle data. And, if you opt to use ASM, you'll probably want to look up what the best practices are for doing so.

Having done consulting for Symantec and VERITAS, I can tell you that NetBackup introduces a whole fun set of considerations - more than can be easily gone into here (if you've got consulting-hours as part of your licensing, you probably want to engage VERITAS for pointers). Things you'll want to consider include: will backups happen over the LAN or the SAN; will you be using backup application-agents or will you be looking to just setup a single, ALL_LOCAL_DRIVES kind of job; will you be using the various accelerator and deduplication options; etc. The mix of technologies you take advantage of can drive what layout options will work best for you.

In general, though, if you're supporting more than one work-load, you'll want to have at least as many LUNs as you do workloads. And, where those workloads have different I/O sub-profiles (e.g., the I/O characteristics for Oracle's redo logs, table-storage and archive logs are all quite different from each other), you'll want to subdivide a given workload's storage in a way that best supports optimizing those sub-profiles.

Close

Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.