Seagate Personal Cloud Retest

[DNet]

I believe Seagate is being deceptive here, and crossing their fingers that no one will notice.

The firmware may have little to do with the performance differences. The big factor is in the drives - in particular, the 5TB drives are SMR, rather than the much faster PMR technology used in the 4TB drives.


For the last month I've been dealing extensively with the 5TB Seagate drives (in both Expansion and Backup Plus guise), and they can be excruciatingly slow - particularly on writes. For example, a straightforward file copy of a few hundred gigabytes or more can start off at over 100MB/s, then around 75GB into the copy begin to slow down dramatically. Eventually - say around 200GB into the copy - the performance has crawled to perhaps 20MB/s or worse. That "3 hour copy job" will now be running all night and into the next day.


Perhaps SmallNetBuilder can ask Seagate for another 5TB unit with the new firmware. While I'm not intimately familiar with the SNB test procedures (and whether or not they run long enough to reveal the performance degredation), in the real world the 5TB model will likely perform much worse than the "replacement" 4TB unit Seagate just provided. Many benchmarks may use (small) data sets which won't reveal the differences, while the average user will experience this painful problem right away as he transfers the data from his old drive(s) to the new one.

 

[thiggins]

I've asked Seagate about this and will post a response when received (if they don't).

Our tests don't run for hours. Largest file copied is a 1 GB VOB file.

And have you run your same test with a Seagate 4TB NAS drive?

 

[DNet]

I've asked Seagate about this and will post a response when received (if they don't).

Our tests don't run for hours. Largest file copied is a 1 GB VOB file.

And have you run your same test with a Seagate 4TB NAS drive?

No, that's just it. It's being run on a fast USB 3.0 UASP connection, and the bottleneck still occurs. Other brand hard drives or RAID arrays (DAS) I'm copying from are not slowing down - they copy from/to each other quickly.

These tests happen to be on Mac Minis using HFS+, but other drives or small RAID0 arrays continue at full speed for an entire Terabyte+ copy of data, regardless of whether on USB 3.0 or Thunderbolt. Based on my experience, the filesystem should not be a major factor here.

4TB WD Green or Red drives, 4TB HGST NAS drives, and 5TB Toshiba drives (all non-SMR) work fine. Though they're inexpensive, since the Seagate external cases already have UASP, I'm fairly confident breaking the drives out of their external cases wouldn't help much.


One thing I have noticed is that very large files (e.g. long video clips) may continue to write quickly, while medium and small sized files (including RAW photos) will slow down dramatically. These are simply sequential copies, mind you.

Sometime in the next few days I plan to archive some long video clips from an NTFS WD RED drive on a Windows 8.1 system to one of these 5TB Seagates (formatted in NTFS this time). I expect they will copy quickly because of the very large file sizes.

 

[thiggins]

I think that's pretty good data. Thanks.

 

[stevech]

medium/small files, of course, with their overhead, diminish greatly the observed throughput, in any storage system.

 

[DNet]

medium/small files, of course, with their overhead, diminish greatly the observed throughput, in any storage system.

I hope you don't mean this to refute my findings. While this may be true in general - and I've certainly seen it in action - this isn't the case here. There's a difference in either this drive or the external case - and I'll bet it's the drive.

As I mentioned, the exact same dataset (comprised mostly of these medium and some small files) copies much faster on the competing drives.

This should give you at least the flavor of what can happen with an SMR drive compared to (now traditional) PMR drives. Note that the 5TB Seagate is in the same family as the 8TB drive tested here:

http://www.storagereview.com/seagate_archive_hdd_review_8tb

With so many different tests - some of which don't show any performance degradation for the SMR drive - some of the facts might get lost in the details. But if you take a good look, you'll see that random writes can be over 50 times slower than traditional drives (!), while some types of sequential writes can be at least 2 or 3 times slower on SMR drives (and perhaps more like 2 to 6 times slower, according to my experience).

 

[stevech]

File System and SMB overhead is fixed - so small size files total transfer w/overhead has to be slower.
Good backup systems run 4 or so streams in parallel - my ADrive and OpenDrive 'net storage services do that. It helps deal with overhead.

 

[DNet]

OK, but I'm saying the design of the Seagate 5TB SMR drive makes it much slower than competing drives (or even Seagate's own 4TB PMR drives). It's dramatically slower in many cases - so much so that the drive's performance can be a bigger factor than anything going on with the NAS itself (which is such an unusual situation these days that one wouldn't normally look for it).

 

[stevech]

Sorry - I missed the "SMR" part. What's SMR?
And PMR?

 

[DNet]

These are methods of writing data to hard disk platters. Perpendicular Magnetic Recording was the previous "new" (now standard) method of increasing platter density. Its growing pain was initial reliability problems. Shingled Magnetic Recording is the latest method. Its growing pain is reduced performance. Because its design frequently requires re-writing data that hasn't even changed, it's possible this drawback will never be overcome sufficiently. So for now (and maybe well into the future) SMR can be acceptable for some archival applications, but may be frustrating in general purpose use.

Take a look at my link in my previous post, or look it up to learn more.

 

[thiggins]

Response from Seagate:

Regarding the Personal Cloud single-bay, your reader is correct in noting that the 5TB is an SMR drive. Although, it is not accurate for him to state that the Firmware has not factored in the performance improvement. The Firmware update has improved the performance on all capacity of drives. As with any new capacity update, you will see variances between drives based on the technology used, but in this case it is not significant enough to have an impact on the customer experience with the Personal Cloud.

Seagate also offered to send another Personal Cloud with a 5 TB drive for another test. I will do this and post results to this thread.

 

[DNet]

Response from Seagate:


Seagate also offered to send another Personal Cloud with a 5 TB drive for another test. I will do this and post results to this thread.

I look forward to seeing the results. As I suggested, it may take a copy of over 100GB (preferably over 150GB) to the Personal Cloud drive for the dramatic slowdown to show up. But I believe my use-case is common (transferring many photos to NAS or DAS), so it's certainly relevant!

Also, is the firmware updated on the drive itself, or only on the NAS device? While I would assume the latter, it's possible to do both (even transparently during a user-applied update). Could you please verify this with Seagate?


Seagate, please note that I never claimed the firmware update has no affect on performance. I merely questioned how much of the improvement was due to it, or whether it's even the predominate factor.


"you will see variances between drives based on the technology used, but in this case it is not significant enough to have an impact on the customer experience with the Personal Cloud" (emphasis mine).

I find Seagate's statement highly dubious - actually, nearly impossible - unless they've updated the drive firmware. I would be happy to validate Seagate's claim if they wish to loan me a 5TB Personal Cloud, or to apply any updates they may have for their 5TB Expansion drives which I own (previous case design that was superseded last month). But when an SMR drive slows to 1/5 the performance of a PMR drive on the first day of use, the customer notices! This characteristic is reflected in some online customer product reviews, as well as in formal reviews such as the one I linked to above.

 

[thiggins]

SNB's NAS benchmarks copy at most ~4GB of data, so won't reveal the alleged behavior.
As agreed, I've passed your contact info to Seagate so that you can sort things out with them directly, should they choose to do so.

 

[sfx2000]

Sorry - I missed the "SMR" part. What's SMR?
And PMR?

SMR and PMR relate to how the data is written to the disk - SMR, at a very high level, is partial overlay of data...

Read performance on an SMR drive should be similar to PMR, but write performance is typically slower on long transfers, unless the OS is aware of SMR.

SMR is probably more useful in a cold-storage environment, or one that is read operations heavy...

StorageReview.com has a nice explanation on SMR in their 8TB Seagate Archive disk review...

 

[DNet]

A rep and a small team of engineers from Seagate contacted me today. They'll try to replicate my results next week. Apparently they used large files (video files, I believe) when they did their large dataset write testing on this drive.

I expect to hear back next Wednesday or Thursday. I don't think they were aware of this issue. It also sounds like it doesn't occur with large files. They're taking the issue seriously - it will be interesting to hear what happens.

 

[sfx2000]

Interesting - keep all advised here...

 

[DNet]

I just got off the phone with Seagate. They confirmed that yes, the issue is very real. It had slipped through their testing. But they also determined this dramatic slowdown only occurs under HFS+ with small/medium files. Apparently this doesn't occur under NTFS, regardless of whether attached to a Windows machine, or on OS X using their Paragon NTFS driver.

I admire Seagate, not only for blazing the SMR trail, but also for taking this condition so seriously. They got many people involved, and are right now working on ways to correct this for future releases. They told me they've sold 3 million SMR drives already, analyzed the returns data, and felt fortunate not too many of those drives have been marketed/sold for Mac/HFS+ systems so far. In other words, most customers are probably getting adequate performance.

We'll see what they do. They may even go so far as to temporarily suspend shipments of SMR drives in their Mac-targeted external drives (e.g. my guess would be some/most 5TB LaCie drives) until they have an update/fix. At least they told me this possibility (not specifically mentioning LaCie) has already been discussed.

They were grateful for the help from the hours I'd put in tracking this down for myself, and said they were able to replicate it with no problem, once they realized it had to be HFS+ with 100GB+ of 'small' files like photos. They're also sending me 6TB drives which shouldn't exhibit this behavior, so I won't have to wait for the fix (I believe these are PMR drives, although they also have 6TB SMR drives).

Finally, all this has got me curious: If I had a 5TB Personal Cloud drive, would it still exhibit this slowdown sometimes? Would the internal filesystem matter as much as how/what protocol a Mac or PC used in attaching to it? After this experience, I feel confident Seagate will address any issues here (if there are any), but I wonder what the current status is?

 

[sfx2000]

I just got off the phone with Seagate. They confirmed that yes, the issue is very real. It had slipped through their testing. But they also determined this dramatic slowdown only occurs under HFS+ with small/medium files. Apparently this doesn't occur under NTFS, regardless of whether attached to a Windows machine, or on OS X using their Paragon NTFS driver.

This doesn't really address Storage Review's comments on the SMR tech... as they replicated results across many test environments...

And I really don't buy the excuse of File System support - sorry sir, that dog just don't hunt...

I think it really comes down to use cases - seems like Seagate has targeted the Archive series more towards ColdStorage/Off-Line storage, and to that point, lower write performance may be acceptable, esp considering the cost and density...

 

[DNet]

Well, I'll have plenty of time to prove you right - or wrong. It's just that I'm not doing my big NTFS transfer just yet. But if the new drives work faster I may be able to do it sooner rather than later ... - when a transfer stretches into day(s) it really holds me up!

I did gather from my conversation that Seagate's been able to implement much of the slowdown-abatement in the firmware, rather than relying on the OS or drivers. At least that's what it sounded like. So perhaps some re-tests will be in order soon - I hope so.


My hunch is that random writes may still get significantly slow at times - probably still unusable for certain applications. But at least sequential writes may be fixed - which theoretically they should be able to do.

 

[stevech]

I don't get it... The drives' transfer rates are already much faster than the local area networks, even 1000BT, due to overhead in TCP/IP, SMB, NTFS, etc. and moreso the file systems.