Sponsors
Sponsor Products
Recommended Workstation Filesystem?
posted by Mike Rochefort  on Nov. 10, 2016, 1:11 p.m. (7 years, 20 days ago)
8 Responses     0 Plus One's     0 Comments  

One of the things that I'm still very novice with is filesystems. I'm looking into transferring a workstation over to CentOS 7, and although I can get myself up and running quickly, I don't know which filesystem is best to choose: Ext 4 or XFS.

Over in the Foundry's community, Ext 4 has been recommended (I always get the warning of using XFS from Mari). I've only ever used used XFS in setting up and trying things out, however. I know the general differences between the two, but for a workstaion running the following apps: Maya, Houdini, Modo, Mari, Substance, Nuke, Fusion, Redshift, RenderMan, with the following storage setup, what would you recommend?

  • 2x 250GB SSD
  • 1x 120GB SSD
  • 3x 1TB HDD (2 of which are  currently software RAID0)

Thanks,

Mike Rochefort


Thread Tags:
  linux 

Response from Mike Rochefort @ Nov. 14, 2016, 11 a.m.

After spending the last hour or so looking into bcache, I've found it pretty hard to locate any good info on CentOS 7/RHEL 7. Todd, you're suggesting that bcache would come in handy if editing over a network, correct? If so, I can understand why using a caching system with a NAS would be worth the setup. Contrary to my previous post, I'm still on the edge about using a caching system for my workstation. The RAID is fast enough as is, and the media drive isn't really accessed too much to begin with. I'm just trying to find something to use the 120GB SSD, maybe dedicate it for Mari projects but those eat space if I remember correctly. It's performance isn't that great either, but it's better than a regular HDD. Formerly, it was an OS X drive for music creation, but I'm trying to refocus myself and dealing with a hackintosh again will just be an added hassle.

Also, I'm fairly new to the 'mailing list' way of doing things, are you guys responding through email? I'm still trying to figure out how to get past posts appended to mine. Sorry for the novice-ness!


0 Plus One's     0 Comments  
   

Response from Todd Smith @ Nov. 14, 2016, 10:35 a.m.
I think this is overcomplicating something you could do easily with four 1TB SSDs in a RAID0.  This would give you better IO thoughput with few IO mechanics to deal with.  Bcache is unnecessary unless you are ultimately looking to apply this to DIY NAS, especially if you are dealing mainly with editing.
Cheers,
Todd SmithHead of Information Technology
soho vfx | 99 Atlantic Ave. Suite 303, Toronto, Ontario M6K 3J8office: (416) 516-7863 fax: (416) 516-9682 web: sohovfx.com

I'm inclined to follow what you said, Bjorn. Looking into LVM cache and bcache and there's a lot of useful information. I noticed a lot more people leaning towards bcache, though. It seems simple enough to setup (relatively speaking), so what kind of configuration or tuning would be needed past that? If it doesn't end up working, I could always just return the drive to its previous state as an OS X drive, but I'd rather not.

I'm considering taking the 2 identical HDD's and re-raiding them into a 2TB RAID0 like I have on Windows. I use this as a scratch/cache disk for apps, it especially comes in handy for simulations. In an effot to remain consistent with what I already use, here's the setup I'm looking at:

250GB SSD = / and /home (coding projects or small media projects)

250GB SSD = Main work drive for projects

1TB HDD     = Main Media drive (where I transfer older projects and host things like movies, etc) Projects will likely get compressed before going here.

2TB RAID0  = Cache disk for all apps and simulations. I would look into NVME drives, but their limited capacity don't warrant the swtich yet.

120GB SSD = bcache disk

A question about bcache: Can I use bcache for both the RAID array and the media drive? Would I have to create 2 partitions on the SSD, one for each backing device? Or should I exclude the RAID from this completely; I usually get upwards of 300MB read/write on it, and I can't think of why it would be read/written non-sequentially. One of the benefits I saw of bcache was that it ignored sequential reads or something like it, which would be super useful when reading media files and could reduce the wear of the SSD. This will be done on a completely clean system, all data will be dumped into dropbox before installation.


To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


0 Plus One's     0 Comments  
   

Response from Mike Rochefort @ Nov. 14, 2016, 10:16 a.m.

I'm inclined to follow what you said, Bjorn. Looking into LVM cache and bcache and there's a lot of useful information. I noticed a lot more people leaning towards bcache, though. It seems simple enough to setup (relatively speaking), so what kind of configuration or tuning would be needed past that? If it doesn't end up working, I could always just return the drive to its previous state as an OS X drive, but I'd rather not.

I'm considering taking the 2 identical HDD's and re-raiding them into a 2TB RAID0 like I have on Windows. I use this as a scratch/cache disk for apps, it especially comes in handy for simulations. In an effot to remain consistent with what I already use, here's the setup I'm looking at:

250GB SSD = / and /home (coding projects or small media projects)

250GB SSD = Main work drive for projects

1TB HDD     = Main Media drive (where I transfer older projects and host things like movies, etc) Projects will likely get compressed before going here.

2TB RAID0  = Cache disk for all apps and simulations. I would look into NVME drives, but their limited capacity don't warrant the swtich yet.

120GB SSD = bcache disk

A question about bcache: Can I use bcache for both the RAID array and the media drive? Would I have to create 2 partitions on the SSD, one for each backing device? Or should I exclude the RAID from this completely; I usually get upwards of 300MB read/write on it, and I can't think of why it would be read/written non-sequentially. One of the benefits I saw of bcache was that it ignored sequential reads or something like it, which would be super useful when reading media files and could reduce the wear of the SSD. This will be done on a completely clean system, all data will be dumped into dropbox before installation.


0 Plus One's     0 Comments  
   

Response from Jeff Yana @ Nov. 12, 2016, 1:22 a.m.

For many years, XFS's performance advantages over other filesystems meant users were willing to tolerate some of the more inconvenient aspects of using it, such as recovering from file system corruption. I am not sure if that is still an issue today. Surprised to learn that its now default on some Linux distros today.


0 Plus One's     0 Comments  
   

Response from Bjorn Leffler @ Nov. 10, 2016, 6:56 p.m.
  • XFS and Ext4 are both very stable and have been for a while (years). One of them have a more robust test suite, which the other (and most other fs) use as well.
  • Looking at public benchmark comparisons, XFS and Ext4 are roughly comparable in performance, if configured for comparable robustness.
  • XFS is better for larger (30T+) file systems.

Mike, in your case, I'd say ext4 and xfs are equivalent. I'd probably:

  • Use ext4.
  • Configure the 1x 120GB SSD as a cache for the larger disks. Bcache works really well, but needs to be tuned for high performance workloads.
    • If you lose the cache disk, then there is no data loss.
    • When you reboot the workstation, the bcache is still hot and active.

0 Plus One's     0 Comments  
   

Response from Mike Rochefort @ Nov. 10, 2016, 5:02 p.m.

XFS being the default was most of the reason why I used it so often. Those changes you speak of in the 4.8 kernel, however  good they may be, are not currently applicable though because RHEL/CentOS 7 run off the 3.10.x kernel. That's good info though! I believe XFS is used more in server space because of it's enlarged drive capacity, which is well over that of Ext 4.

Would you recommend formatting all the drives to Ext 4? I usually mix things up, on Windows I have NFTS for boot, HFS+ on 2 other drives (dual-booting Hackintosh), exFAT on my work SSD, and NFTS for the RAID0 drives. I'm simply not aware enough to think of whether there's a benefit to doing this or not.

- Mike


0 Plus One's     0 Comments  
   

Response from Jean-Francois Panisset @ Nov. 10, 2016, 4:45 p.m.
XFS is the default for RHEL / CentOS 7, so the "principle of least surprise" might be an argument in favor of that.

Also XFS is still getting some significant development. In the 4.8 kernel the "reverse mapping" infrastructure went in:

https://lkml.org/lkml/2016/8/6/104

==
This reverse mapping infrastructure is the building block of several upcoming features - reflink, copy-on-write data, dedupe, online metadata and data scrubbing, highly accurate bad sector/data loss reporting to users, and significantly improved reconstruction of damaged and corrupted filesystems.
==

although these require changes to the on-disk format, so probably not immediately relevant to how you are building CentOS 7 workstations...

Now if I could only find my installation CD for IRIX 5.3+XFS...

JF


On Thu, Nov 10, 2016 at 1:25 PM, Jesse C <crimson.corelio@gmail.com> wrote:
I use ext4 on all my stuff since it is also readable via windows in a pinch. I've never had any bad experiences with it, so I've never felt the need to look at xfs.

To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


0 Plus One's     0 Comments  
   

Response from Jesse Clemens @ Nov. 10, 2016, 4:30 p.m.
I use ext4 on all my stuff since it is also readable via windows in a pinch. I've never had any bad experiences with it, so I've never felt the need to look at xfs.

0 Plus One's     0 Comments