Throughout my experience reading through the forums, there are a few chosen individuals that always amaze me with their vast scope of Linux/unraid knowledge and they are always willing to share it. I never want to miss an opportunity to thank everybody for the help that they bring. One thing that I seem to have noticed though, is that it seems like the majority of these veteran users seem to not use the user shares feature.  I'm curious as to why? Is it an issue with performance, compatibility, or something else entirely?

I'm a complete noob, and just ran into a problem with User Shares.

I have a:

1.5tb parity

1.5tb data 1

1.5tb data 2

500gb data 3

500gb data 4

1.5tb data 5 that I just put in so there's nothing on it

 

I made a "storage" share and I copied all my stuff to the unraid to this "storage" share.. set at split level 2 with file allocation of HIGH WATER.

my first 1.5tb had like only 600megs free, I went to copy about 3 gigs of stuff to the "storage" share and well it auto put it on disk 1 and told me it is out of space!

I had to manually navigate to disk5 and manually made directories/folders to match and copy it there.

 

Now, after copying some more files, my data 1 drive only has 1.8 MEGS free, and I go to copy a 15 meg file onto the "storage" share and guess what? it like crashes and won't copy.

 

Is this a normal problem with "high water" file allocation?

 

I wanted to just make 1 share and have my unraid act as 1 giant hard drive.. but looks like it's not going to work that way. sigh.

 

EDIT: after messing around a bit, the problem may be with using TeraCopy (to copy files over), using regular windows copy/move seems to be fine.

One thing that I seem to have noticed though, is that it seems like the majority of these veteran users seem to not use the user shares feature.  I'm curious as to why? Is it an issue with performance, compatibility, or something else entirely?

User-shares did not get introduced until version 4.2 of unRAID.  (17 versions of unRAID from the original version supplied by Tom)

Many of us "old-timers" had our servers well organized before that feature was added.

 

I use user-shares to present a consolidated view of my movies to my media players.  But... I normally write directly to a specific disk using the hidden, but writable "disk shares"   I don't use "split-level" to keep the files affiliated with a movie together, as I use ISO images... (one file per movie)

 

I've had my server since the very first 1.050826 release (the August 26, 2005 release)  Back then it only supported 12 IDE disks (11 data).  I started with a pair of 500 Gig drives... they were the biggest disks available back then... and about $350+ each.   I did not have to worry about which disk share.... I only had one.  It was "disk1' 

 

So, to answer your question... it has nothing to do with performance or compatibility.  It has more to do with "control" of where we put our data.  Oh yes, not sure about today, but last time I tested, accessing through a "user-share" was slower than accessing a disk directly.

 

I just made a quick test, reading from both the disk directly, and through the user-file-system.   I got 36 MB/s reading directly from the disk, and 33MB/s from the user-file-system... just 3MB/s slower...  at one point in the past, there was almost a 50% difference in performance.... I think that was before Tom completely re-wrote the user-file-system.

 

Directly from the disk

root@Tower:/boot# dd if=/mnt/disk8/Movies/JAMES_BOND-2009-QUANTUM_OF_SOLACE.ISO of=/dev/null bs=16k

387982+1 records in

387982+1 records out

6356699136 bytes (6.4 GB) copied, 175.272 s, 36.3 MB/s

 

root@Tower:/boot# dd if=/mnt/disk8/Movies/JAMES_BOND-2009-QUANTUM_OF_SOLACE.ISO of=/dev/null bs=16k

387982+1 records in

387982+1 records out

6356699136 bytes (6.4 GB) copied, 176.391 s, 36.0 MB/s

Through the user file system

root@Tower:/boot# dd if=/mnt/user/Movies/JAMES_BOND-2009-QUANTUM_OF_SOLACE.ISO of=/dev/null bs=16k

387982+1 records in

387982+1 records out

6356699136 bytes (6.4 GB) copied, 190.096 s, 33.4 MB/s

 

root@Tower:/boot# dd if=/mnt/user/Movies/JAMES_BOND-2009-QUANTUM_OF_SOLACE.ISO of=/dev/null bs=16k

387982+1 records in

387982+1 records out

6356699136 bytes (6.4 GB) copied, 188.169 s, 33.8 MB/s

 

 

Joe L.

One thing that I seem to have noticed though, is that it seems like the majority of these veteran users seem to not use the user shares feature.  I'm curious as to why? Is it an issue with performance, compatibility, or something else entirely?

 

I have user shares enabled, I do not use them often,  as it causes too many disks to spin up.

Also I like organizing my files and knowing where everything is.

In the event of a disk failure, I'll have an idea of what is in jeopardy.

The fact remains, with rtorrent running & seeding a number of torrents, 8GB of ram is still not enough to keep disks from spinning up.

There are just too many files and too much activity to keep disks from spinning up.

I'll probably end up moving my rtorrent process to a smaller machine so that my unRAID server is more idle.

In addition, my rsync backup process uses the rsync --link-dest methodology.

This lets me rsync only files newer then the latest backup.

One of the issues is the user share filesystem does not support hard links.

Furthermore, The rsync backup process increases the number of files as each machine's files are duplicated to the backup disk.

 

I usually run a daily mlocate on the system.

This logs where every file is and provides a command line utility to do a search down the whole db in mere seconds without spinning up a disk.

So If I want to watch a movie, I login, do a locate, spin that disk up, go and watch the movie.

Not the most efficient. I'm hopeful that later versions of unRAID will have a real web server so I can write CGI's to handle the locate and spin up from a browser. We'll see.

All interesting perspectives. Thank you.  I asked the question out of the suspicion that there was a larger discrepancy between disk shares and user shares for more users.  I had always been moderately happy with my unraid performance until I started noticing stuttering streaming HD content. My testing seems to show that at least for me, there still exists a huge discrepancy in disk share vs user share.

 

Posted by Joe L.

I just made a quick test, reading from both the disk directly, and through the user-file-system.   I got 36 MB/s reading directly from the disk, and 33MB/s from the user-file-system... just 3MB/s slower...   at one point in the past, there was almost a 50% difference in performance.... I think that was before Tom completely re-wrote the user-file-system.

 

My Testing

 

Directly from disk

root@Bastet:/mnt/disk1/Movies/Wanted (2008)# dd if=/mnt/disk1/Movies/"Wanted (2008)"/"Wanted (2008) BluRay 1080p.mkv" of=/dev/null

16659512+1 records in

16659512+1 records out

8529670427 bytes (8.5 GB) copied, 140.783 s, 60.6 MB/s

 

Through the user file system

root@Bastet:/mnt/disk1/Movies/Wanted (2008)# dd if=/mnt/user/Movies/"Wanted (2008)"/"Wanted (2008) BluRay 1080p.mkv" of=/dev/null

16659512+1 records in

16659512+1 records out

8529670427 bytes (8.5 GB) copied, 142.45 s, 59.9 MB/s

 

So indeed I have similar results, but when this same file is streamed across the network I see the following:

 

Unraid -> Ubuntu Box SMB disk share 27.6MB/s

Unraid -> Ubuntu Box SMB user share 12.1MB/s

 

Unraid -> Ubuntu Box NFS disk share 49.1 MB/s

Unraid -> Ubuntu Box NFS user share 17.4 MB/s

 

I too take advantage of user shares primarily to provide a consolidated view of my collection to my media centre, but the discrepancies in transfer speeds are puzzling.

 

Anybody have any thoughts, suggestions, or even similar experiences with user shares?

 

 

 

 

 

I use user shares for the same reason; consolidating the view for my modded xbox running xbmc.

 

I do do it slightly different then some though.  I use the user share to write to also, but I restrict my disks so that I can control which ones get filled with which content.  I believe there is a new feature that is "fill in order" that essentially does what I want it to, but I have not updated to new 4.5 beta as there are some other stuff that I am running that does not work in 4.5

 

Directly from disk

root@Bastet:/mnt/disk1/Movies/Wanted (2008)# dd if=/mnt/disk1/Movies/"Wanted (2008)"/"Wanted (2008) BluRay 1080p.mkv" of=/dev/null

16659512+1 records in

16659512+1 records out

8529670427 bytes (8.5 GB) copied, 140.783 s, 60.6 MB/s

 

Through the user file system

root@Bastet:/mnt/disk1/Movies/Wanted (2008)# dd if=/mnt/user/Movies/"Wanted (2008)"/"Wanted (2008) BluRay 1080p.mkv" of=/dev/null

16659512+1 records in

16659512+1 records out

8529670427 bytes (8.5 GB) copied, 142.45 s, 59.9 MB/s

 

So indeed I have similar results, but when this same file is streamed across the network I see the following:

 

Unraid -> Ubuntu Box SMB disk share 27.6MB/s

Unraid -> Ubuntu Box SMB user share 12.1MB/s

 

Unraid -> Ubuntu Box NFS disk share 49.1 MB/s

Unraid -> Ubuntu Box NFS user share 17.4 MB/s

Yery interesting...

I did not expect to see that huge difference at all when used over the LAN.

About all I can think of is a buffer size issue of some kind being involved.  Or a concurrency/blocking issue with the network card driver...  (can't be SAMBA, or NFS, since neither has anything to do with the other.)  The only common point is the networking ... and how it interacts with the user-share vs. the /dev/mdX file-system.

 

Joe L.

Just to to re validate my results, I performed the same tests today, as well as the tests on my second unraid server, different motherboard, with different NIC and was met with similar results

 

Unraid2box -> Ubuntu Box SMB disk share 33.5MB/s

Unraid2box -> Ubuntu Box SMB user share 26.MB/s

 

Unraid2box -> Ubuntu Box NFS disk share 51.1 MB/s

Unraid2box -> Ubuntu Box NFS user share 24.7 MB/s

 

I also ran the tests with 512MB of system memory, and then re ran the tests with 2.0GB of system memory to see if memory was an issue, but I achieved the same results each time the test was done. Ping tests show my latency between boxes averages @ 0.065 ms. I installed Iperf on the Ubuntu client and both UnRaid boxes and my network caps at a steady 980 Mb/s.

 

Joe. L. wrote

I did not expect to see that huge difference at all when used over the LAN.

About all I can think of is a buffer size issue of some kind being involved.  Or a concurrency/blocking issue with the network card driver...  (can't be SAMBA, or NFS, since neither has anything to do with the other.)  The only common point is the networking ... and how it interacts with the user-share vs. the /dev/mdX file-system.

 

 

I'm under the impression that I've eliminated my hardware and my network from the equation as well as networking protocols. This leaves unRAID as the culprit. So where do I go from here? Are there any other options that other users have tuned to minimize the user share performance discrepancy? Is this something that perhaps Tom would be able to comment on?

 

As you can see I've put in a bit of effort in testing to eliminate my issues once and for all, so I'm definitely open to suggestions!

Is this something that perhaps Tom would be able to comment on?

I thank you for your testing efforts...  I've sent an e-mail to Tom pointing him to this thread.

 

Joe L.

Tom responded to my message regarding the difference in "read" speed when reading from a disk vs. user-file system over the network.

 

He thanks us for the analysis, and he did say he has noticed a difference himself, had not yet looked into it.  Other items were higher in priority, as even the slower rate of 12 MB/s was fast enough for most HD material.   He said he has not noticed any issue with his own HTPC on HD material over SMB, but realizes it might show itself on some material on some players.

 

Joe L.

I thought I would respond to the OP question.  When I first starting using unRAID, the user share file system was relatively new and users were reporting various issues - including performance.  Also, the initial implementation would not allow you to write to a disk share and update the user share automatically.  Despite some improvements, the user share feature continues to have performance issues and is still the most problematic feature of unRAID, especially for new users.  The whole "split level" concept is perhaps a necessary evil, but seems to confuse each and every user that tries to make it work.

 

I like the flexibility to custom place my content, so would never write to a user share.  And I have my HTPC player configured so I can browse my entire library without spinning up any disks, and when I select something to play, it only spins up the needed disk (and I don't have to run the constant directory scans).

 

I do have a custom Samba configuration that lets me set up my own share names and implement security, with all (plus more) flexibility than provided by user shares.  And I read and write using these meaningful share names and never use the cryptic "disk1, disk2" shares.

 

The only reason I would use user shares if I used some player that forced me into it - and then it would only be for reading.  For now though, I just don't need it.

 

I would advise anyone working ot set up their server to NOT use user shares for the first month.  If, after that, you feel you need it, turn on the feature then.  At least you'll know the array is loaded properly and at full speed.  And you'll learn some of the ins and outs of unRAID before tackling this difficult feature.

Joe L. wrote

Tom responded to my message regarding the difference in "read" speed when reading from a disk vs. user-file system over the network.

 

He thanks us for the analysis, and he did say he has noticed a difference himself, had not yet looked into it.  Other items were higher in priority, as even the slower rate of 12 MB/s was fast enough for most HD material.  He said he has not noticed any issue with his own HTPC on HD material over SMB, but realizes it might show itself on some material on some players.

.

 

Thank you for keeping me up to date on this.  I didn't know what to expect from Tom, and I must admit that I am a little dissapointed that user share performace wouldn't rank as a high priority. True that 12MB/s might be passable for most HD material, but to give the impression that this speed is "good enough" kind of lowers unRAID as a product targeted towards end users.

 

Ideally my intent for an unRAID box is to provide a redundant, centralized storage pool that's data can easily be consolidated and exported throughout the household. However it seems that performace limitations will prevent me from watching a BluRay backup on my big screen while the girlfriend watches 720p episodes of her favorite show in the bedroom if I choose to enable user shares. I would think that this wouldn't be too high an expectation. I will look into resetting up my xbmc libraries via 12 exported disk shares rather than 1 Movies share. Frustrating? Definiately.  I just hate the feeling that when solving a problem the solution lies in working around the problem rather than solving the underlying issue.

 

It just brings me back to the focus of why I started the thread and asking about why people configure there system in different ways. There's always a different way to approach things, and the posts here have been very informative.

 

I just hope that in the future Tom will be able to put some time and effort into making issues such as mine a priority. I love my unRAID box, and will undoubtably continue to do so. Unfortunately, when I recommend unRAID to friends they're always skeptical of the ease of use of such a system despite my demonstrations of how easy it is to use and administer. When features such as user shares increase in complexity and decrease performance, it makes the task of convincing friends how awesome unRAID is even harder.

I can't say I'm an experienced UnRAID user (< 1 year) but I've been playing around with htpc's for quite some time and always been looking for a full control on every component. With UnRAID this means using the disk shares to write and user shares when reading from a htpc-clients. Disk shares give much better performance (see below) and control where the data is located but user shares enable zero configuration on client side when adding new disks to the server. I have the user shares exported as read-only for safety purposes. My setup consists of Meedio combined with DVD Profiler integration with XML2Mee which works perfectly now that I've Cachedirs running on UnRAID; disks stay spun down when Meedio refreshes it's media library. So pretty much the same reasons as the others have listed.

 

For the performance between disk and user shares I made quite comprehensive testing when building the system. The whole story is here (all the graphs are linked):

http://lime-technology.com/forum/index.php?topic=3958.0

And the final conclusions here:

http://lime-technology.com/forum/index.php?topic=3958.0#post_anch5

 

Basically what I found that user share performance was about 60% of disk share performance. In absolute values server internal read speed was 70MB/s (user) vs 115MB/s (disk) and write speed 30MB/s (user) vs 45MB/s (disk). For the cache drive read speed was 120MB/s and write speed 100MB/s. I have yet to finish (or even start) client side testing. As I currently have only two htpc-client I haven't been able to perform any true multi client tests.

 

Now if you compare these values to a current Blu-ray titles having a bit rate normally less than 50Mb/s (=6,25MB/s) even with smb/windows/network/concurrent usage or whatever slowdowns you could be hosting quite many Blu-rays.