I've got a new system (Ubuntu 14.04 on an SSD), including 2*Megaraid 9271-8i controller+cachecade & BBU connected to the backplanes of the case. Each controller has 8 WD RED 3TB (4096 Physical block size) connected via these backplanes. I've created 2 Raid6 LVs with 256kb Stripe size.

During my reseach for the optimal FS type and mounting options some questions are still unansered:

  1. Should I partition the Raid-Drives with one large partition and align the partition, according to the block size of the HDs and stripe size? Or should I put the FS on the plain raid6 LV device?

  2. Which FS should I use? I've planned to use ext4, as I'm completely unfamiliar with xfs. What is your recommendation?

Performance is a crucial point here, as it can happen that several clients are reading and writing to these Raid disks. ZFS is not an option due to missing ECC RAM. I've not planned to extend the amount of Disks per Controller. I'm storing mainly large files on this storage, threfore I've used a rather large stripe size, compared to what I've seen in some guides.

If I should partition the full disk, which paramters should I use to align the partition accordingly with parted?

And which parameters should I use to create and mount the FS? For ext4 it would be mkfs.ext4 -m 0 -b 4096 -E stripe-width=384,stride=64 for the the creation and nosuid,nodev,nodiratime,nobarrier,noatime,nofail, 0 0 for mounting

Have I missed somthing and what should I use in case of xfs

Best regards

  • Why wouldn't you use ECC RAM? Is this server-class hardware? – ewwhite Feb 14 '15 at 16:55
  • nope.. rest of the HW: i7-4820k, 32GB G.Skill TridentX, ASRock X79 Extreme9. The "server" is used to run multiple VMs in paralell as well on the SSD. And I have the option to add a 3rd Megaraid card, to fill up the rest of the 24-HD-bay. – Nighter Feb 14 '15 at 16:57

1) Assuming that you boot from some other disk you might just as well put the file system on the raid disk without any partition table. This is if you prefer single big file systems filling up the entire raid disks. If you for some reason want to split up the raid you should of course use a partition table for that.

2) If you feel comfortable with ext4, stick to ext4. I have used reiserfs, xfs and ext4 on raid systems myself. Among those I would have preferred reiserfs a few years ago, but today I prefer ext4 as reiserfs now seems to suffer from some bit rot.

Before putting a file system on the raid system I would do some benchmarks of raw disk performance and compare different stripe sizes. Looking at the results of those benchmarks you will most likely find that the network will become the bottleneck for your clients. This will probably also be true for big files, but for small files the file system will become the bottleneck. But there is no big difference in performance between different file systems, they all get worse performance for small files.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.