I want to use unraid for my new server box, and I got a couple of Samsung F4 drives from newegg the other week. I only have movies that I will write once to the drive, but I'm not too sure I want to use unraid with them. Does anyone know if unraid will be supporting 4K drives in the near future. I really want to start using them. Otherwise I'll have to look at other software.

 

Thanks

It better do it soon. Linux generally already does so unraid can do it but Limetech needs to make it happen.

FYI, the 4k sectors do not appear to the OS or the SATA port.

 

The problem is the sector alignment at sector 63, not the 4k sectors. If unRAID is changed to start at say sector 64 then it breaks all the existing drives. In that manner, WD had a good idea providing the jumper to shift the alignment.

 

The drive wil still work as a data drive for media files. It would be a poor choice for a parity drive or a data drive where you are storing lots of small files.

 

Peter

 

FYI, the 4k sectors do not appear to the OS or the SATA port.

 

The problem is the sector alignment at sector 63, not the 4k sectors. If unRAID is changed to start at say sector 64 then it breaks all the existing drives. In that manner, WD had a good idea providing the jumper to shift the alignment.

 

The drive wil still work as a data drive for media files. It would be a poor choice for a parity drive or a data drive where you are storing lots of small files.

 

Peter

 

If I do it for media files and don't care about write performance, will it still be okay for me? Data integrity wise, is it the same as using any other hdd?

FYI, the 4k sectors do not appear to the OS or the SATA port.

 

The problem is the sector alignment at sector 63, not the 4k sectors.

 

Correct. So far no drives expose 4K sectors to the SATA controllers or OS.

It will work for you.

 

I also will say it will work exactly the same as a parity drive OR as a data drive.  

The parity drive's sectors that are written are exactly the same sectors that are written in the data drives.  

 

Now both parity and data will be less efficient using this drive on multiple small files in a performance benchmark, but on media files it probably will make little difference.  Additionally, unRAID reads and writes in blocks of data far larger than 4k, so having to read one more block at the end will probably not be noticed by most.   (notice how many posts say something like "I did not jumper the drives, and they've been working for months")

 

Joe L.

FYI, the 4k sectors do not appear to the OS or the SATA port.

Yes, but the problem is the drive will have severely degraded performance if the OS doesn't partition it correctly to use 4K sectors (or so I have heard). So the OPs question still stands. When will unraid support 4k drives? I think adding a simple 4k checkbox to the devices tab or when you tell unraid to format the drive should work. It will have to support it eventually as it looks like all drives will eventually be this way.

FYI, the 4k sectors do not appear to the OS or the SATA port.

Yes, but the problem is the drive will have severely degraded performance if the OS doesn't partition it correctly to use 4K sectors (or so I have heard).

You've heard incorrectly.

So the OPs question still stands. When will unraid support 4k drives?

I do not expect it to be soon.  Possibly after 5.0 is out of beta, and that has been almost a year in the making.  Only lime-tech knows.... and they have not said anything about it at all.    I'd not expect it in a few months... not if history is any guide.  If you absolutely need it, unRAID is not your solution.

I think adding a simple 4k checkbox to the devices tab or when you tell unraid to format the drive should work.

It will not do it.  It is FAR more complex than that.  The reiserfs file system that unRAID uses already uses 4k blocks as its default size.

It will have to support it eventually as it looks like all drives will eventually be this way.

It will eventually have to support drives larger than 2TB.  That is a MUCH bigger issue, but related.   The 4k drives today still present 512 bytes sectors to the OS.  Nothing needs to be changed to use them as long as they are not bigger than 2TB.

 

Joe L.

You've heard incorrectly.

So if I un-jumper my 4k drives and repartition them they will perform just the same as if they were jumpered? What's the point of the jumper then if there is no performance issue?

Yes, but the problem is the drive will have severely degraded performance if the OS doesn't partition it correctly to use 4K sectors (or so I have heard).

 

There can be a performance hit when reading many small files but for large files it make little difference. If you got a good deal on the drive or only trust Samsung drives then use them. A 1meg files means somewhere around 250 sectors are read or written. Do you really think it'll make much speed difference if 251 sectors are read or written instead? 1 gig is 250,000 vs 250,001 sectors.

 

If you are reading files in small bits (say during media playback) then the drive might be reading many more sectors overall but it's spread out over time so there is no perceived performance loss. In other words, the drive might be reading a fair number more sectors but you are expecting the data so slowly that it makes no difference.

 

Peter

 

You've heard incorrectly.

So if I un-jumper my 4k drives and repartition them they will perform just the same as if they were jumpered? What's the point of the jumper then if there is no performance issue?

I did not say there would be no performance difference.  I said you would probably not notice the difference.

 

For example many people are saying performance is bad if writing lots of small files.  Perhaps, but does it affect you?  If I write a two block file, and it takes 3 revolutions of the disk to write it if perfectly aligned, and 6 revolutions of the disk if not, let's see...

 

7200 RPM = 120 revolutions per second =  .008333 seconds per revolution.  

For my two examples it would then take .025 seconds more in the worst case.   Would I notice the extra 2.5 hundredths of second when saving my file.  Not likely.   Lets take a larger file.  Lets assume it is 100 blocks long.  Let's assume we write one block at a time (worst case, as Tom has actually stated it writes more like 128k at a time) Now, at one block at a time we only need to add 1 additional block for the last block that traverses a 4k boundary.

 

So, we write 100 blocks in a perfectly aligned system vs. 101 in an imperfectly aligned system.   .8333 seconds vs. .8416 seconds.   Will I notice the extra .00833 seconds... not likely.

 

Do not get frightened off by sector alignment.   Yes, if you write thousands of tiny files in quick succession by a disk performance test, it will be measurably slower... but does it affect your use of the server, and it is how you use your server?

If you are reading files in small bits (say during media playback) then the drive might be reading many more sectors overall but it's spread out over time so there is no perceived performance loss. In other words, the drive might be reading a fair number more sectors but you are expecting the data so slowly that it makes no difference.

Actually, in reading files during playback the memory buffer in the disk drive probably caches the whole track.  It almost certainly performs read-ahead, (and buffers the subsequent blocks).  I doubt sector alignment is an issue at all in reading files that were mostly saved as large numbers of contiguous blocks.

Here's how unaligned drives (WD EARS) will perform on file writes versus aligned drives in very typical usage (test performed copying 1.5-2GB ISO image -- http://linuxconfig.org/linux-wd-ears-advanced-format)

 

Unaligned

reiserfs: 87 MB/s

ext2: 114 MB/s

ext3: 47 MB/s

ext4: 92 MB/s

 

Aligned

reiserfs: 101 MB/s

ext2: 126 MB/s

ext3: 87 MB/s

ext4: 106 MB/s

 

I imagine the Samsung F4 performance degradation is similar.

 

 

Despite what others are saying, the issue looks to show up when writing larger files too.

Here's how unaligned drives (WD EARS) will perform on file writes versus aligned drives in very typical usage (test performed copying 1.5-2GB ISO image -- http://linuxconfig.org/linux-wd-ears-advanced-format)

 

Unaligned

reiserfs: 87 MB/s

ext2: 114 MB/s

ext3: 47 MB/s

ext4: 92 MB/s

 

Aligned

reiserfs: 101 MB/s

ext2: 126 MB/s

ext3: 87 MB/s

ext4: 106 MB/s

 

I imagine the Samsung F4 performance degradation is similar.

 

 

Despite what others are saying, the issue looks to show up when writing larger files too.

Those were not unRAID measurements... They do not have the same read-rotate-write-rotate-read-rotate-write-rotate pattern that unRAID would use.

 

Sorry... but not a benchmark you can use for a direct comparison.  Once you factor in rotational latency and track buffering, the difference may be less noticeable in a media server than you think.

Sheesh, you sound like a politician.

 

The point is, it would be better if unraid started supporting them. Also, wont it eventually HAVE to in the future? Or are hard disks going to have emulation/translation firmware until eternity?

Sheesh, you sound like a politician.

I can't be...  I'm too honest. 

The point is, it would be better if unraid started supporting them. Also, wont it eventually HAVE to in the future? Or are hard disks going to have emulation/translation firmware until eternity?

how long have we had the current MBR partitioning scheme?  And MS-DOS three character extensions on files?

 

The need to emulate older drives will remain for a very very very long time.  Remember, basically every PC out there is not compatible.

 

Yes, unRAID will need to support them, but it is not an immediate (within 2 months) need.

Geeze, once again, all drives appear to be 512b sectors to the outside world, even the new 3T drives that have been released. Despite one of the posts in the links below, until I see a confirmed screen capture I won't believe otherwise. So, as of today, every drive is compatible. Talking about having to support 4k sectors ~natively~ (ie no  drive translation) is fairly useless when talking about needing to support this drive.

 

The problem, which has been grossly blown out of proportion for most typical PC use, is having a partition that does not align to the 4k sectors on the drive. In other words, the first 4 512b sectors do not land on a common 4k sector internally to the drive. So, if you read between 512b and 4k of data you could potentially require the drive to read 8k of data (it would need to read 2 x 4k sectors). But then, even if reading 4k of data who says that 4k of data will just happen to be aligned itself to the drive sectors when written? It would be completely random if this happened or not. Remember, you are writing 512b sectors and the drive is internally sticking those 512b into a 4k sector. There is no reason that writing 4 x 512b sectors would have to line up with a drive 4k sector.

 

So, in the real world what does this really mean? Who has a whole bunch of 4k files these days? Who even has many 512k files?

 

Here are some links;

 

http://lime-technology.com/forum/index.php?topic=7845.0

 

It doesn't seem to have much issue with any reads and large writes when looking at the tests here.

 

http://lime-technology.com/forum/index.php?topic=8012.0

 

Peter