Kevin OBrien

Seagate Archive HDD Review (8TB) Discussion

Recommended Posts

Since there is a lot of vagueness around smr I would say explore it further. Both in hardware and software solutions like BTRFS and ZFS along with random writes but also initial writes of an array/pool. I mean when you repopulate an array with a large copy instead of a rebuild how would that affect performance for instance since it's basically a very long sequential write.

Then again I'm not an expert here just a home user with a large data volume interested in those 2 things with mainly just large files.

Share this post


Link to post
Share on other sites

Well I figure I'd add my experience of using these on ZFS seeing as the last poster just asked :)

So far: Inconsistent sustained write performance but no issues using it in ZFS RAIDZ Currently touching on 5TB of sustained writes to each drive, they are still maintaining full speed averaging in the region of 100MB/sec each in the array but bursting to around 200MB/sec regularly. I would say the average is actually considerably lower than the true average would be if these were not used in a RAID as all drives get held up during the short 1-2 second dips when one drive slows down.

Basically, the sustained I/O profile after 5TB of non-stop writes to an 8TB drive is something like:

180-195-180-170-49-170-201-195-130-195-190-200-195 (MB/sec) etc.

Conditions are 256KB requests, queue size 16, average response time remains reasonable at 12ms.

That said, this is the pattern I've been seeing since first using these disks, well before the initial 20GB landing zone could have been filled - suggesting the idea large writes with write-back caching go straight to disk and skip the landing zone.

Furthermore, I suspect the dips in performance every 5 seconds or so are due to ZFS txg syncs, which are a number of small (<=128KB) writes that cannot be cached.

While I've only had these drives a couple of days it seems that in single-user scenarios and light-duty NAS tasks they work just fine. Furthermore, they're outperforming the Seagate 4TB NAS drives by a good 40-50% in sustained RAIDZ writes.

(The test array is 5xST8000AS0002 drives in RAID-Z(1) on a Xeon E3 v3 system. There were just over 21TB writes to the array, around 5TB to each drive. Comparison NAS drives were ST4000VN000)

Oh, and also getting around 180 IOPS in 8K random reads, QD2, 7.5ms latency.

One thing I ought to add. Contrary to traditional RAID systems, ZFS can almost be thought of as a hybrid object-based storage system when dealing with small files and metadata. With large sustained writes like the above it behaves much the same as standard RAID5/6/etc. Another important factor being it's copy-on-write (never overwrite in-place) system, which would work hugely to the advantage of SMR drives if only they also supported TRM (which they do not - yet).

Share this post


Link to post
Share on other sites

What are the sustained sequential write stats for this drive? I don't remember seeing any in the review. I would have thought that it would bypass the cache for such writes, writing them directly to the shingled portion of the drive (and the Skylight report certainly implied that this was a likely strategy).

It does seem to bypass - as per my uncontrolled tests, these drives will sustain at least 100MB/sec for over 5TB of continuous writes, probably more. Peak STR writes are around 190-200MB/sec at least as far as 5TB into the drive. The HD Tune graph posted earlier seems to be pretty accurate.

Once I get round to it I'll pull a drive off my RAID-Z and force a clean rebuild to see how it performs.

The landing zone is 20GB correct. And again as mentioned the synthetic test didn't show the sustained sequential drop but the single drive Veeam and separate RAID1 rebuild figures did. No magic there, all HDD vendors know this and will tell you as much with SMR drives.

SMR handle both random and sequential bursts as sequential writes (hence the high 4k burst figure). Once it leaves the landing zone all bets are off.

It'd be interesting to see if zero-filling a drive would "reset" it's sustained write performance to the levels I saw. It seems to me the behaviour we're seeing with these early SMR drives is very similar to that of early (pre-trim) SSDs. The arguments against RAID'ing them apply similarly to the arguments against RAID'ing SSDs.

Anyone know what is new with V2? I really need at least 8TB drives. I only have room for 12 drives in my system and no physical space for a second array. The HE drives are out of my budget. I'm hoping V2 fixes the problems with RAID but I'm not betting on it.

We haven't had any outreach on a new gen of Smr.

I suspect he's referring to the drive(s) you tested, which are marketed by Seagate as Archive HDD V2. The V1 being the first model that topped out at 5TB.

That HDTune benchmark is also somewhat interesting: your test shows a maximum sequential write speed of 190 MB/s, which matches the read speed and the data sheet. However, overclockers.at show a constant write speed of 150 MB/s over a large capacity range. This is unphysical for a HDD, except if it was limited by the interface (it's not). Or, as Brian said, the drive would actually be writing to the reserved landing zone all the time. Can we conclude form this that the landing zone is not at the beginning of the platters, but further in? At the position where STR drop to ~150 MB/s. Looking at the read STR graph from overclockers.at show this to be the case around 4.6 TB, or "roughly in the middle" of the platters. It also matches the average read speed of 152 MB/s very well. Apparently Seagate choose to optimize average acess times by placing the landing zone in the middle, while sacrificing 40 MB/s write speed. Sounds like a good general purpose trade-off.

I think you're overthinking it. If you ask me, the 150MB/sec on their write charts is simply an error in their testing. A quick and dirty ATTO test I did yesterday showed write speeds of 180MB/sec+ and very similar to what I saw during ~5TB of sustained writes.

Furthermore: at 6k rpm (according to overclockers.at) a maximum speed of 190 MB/s is quite large and way above what drives with 1 TB platters can achieve. This suggests that without shingling the drive would have 1.1 - 1.2 TB platters. Or where else should this performance come from? Shingling "only" increases the track density by overlapping them, but doesn't change the linear density, does it?

This drive has 1.33TB platters.

@Kevin: do you know what happens during that raid rebuild? Is the content of the valid drive simply copied to the new one, or are the contents of both compared and only differences are resolved?

During a standard RAID rebuild it is purely 100% writes of data from the other drives. During some edge cases on ZFS it can be involve verifying existing data - but AFAIK no traditional RAID (hardware or software) does that.

Share this post


Link to post
Share on other sites

I think you're overthinking it. If you ask me, the 150MB/sec on their write charts is simply an error in their testing. A quick and dirty ATTO test I did yesterday showed write speeds of 180MB/sec+ and very similar to what I saw during ~5TB of sustained writes.

This drive has 1.33TB platters.

@point1: Well, matching numbers for one theory do not automatically mean it's true :D

@point2: yes, 1.33 TB platters including the capacity boost from shingling. I'd like to know how large this boost is. Based on traditional 1 TB platters it would be 33%. I think Seagate claimed around 20%, which would match my assumption of 1.1 - 1.2 TB platters without shingling.

And thanks for your other information!

MrS

Share this post


Link to post
Share on other sites

@point1: Well, matching numbers for one theory do not automatically mean it's true :D

While I myself tend to argue 'the plural of anecdote is not data', in this case we have three tests indicating a sustained write speed of >180MB/sec and only one test indicating 150MB/sec. Personally, I'm also biased towards preferring my own data, but even without it, the quantity of evidence suggesting 180+ outweighs the quantity suggesting 150 :)

Unfortunately I'm yet to discover anything on my system that can consistently generate enough random data to continuously max out a drive, /dev/urandom only produces 15MB/sec for me, and /dev/zero may be unreliable for testing drives with internal garbage collection.

That said, did a ZFS rebuild the other day as well as re-create and re-write another 5TB of data to each drive and saw no decrease in performance on the 'dirty' drive, *except* ironically, during the first 20GB which appeared limited to ~50MB/sec. After the first 20GB the remaining writes jumped up to the normal 100-180MB/sec range I saw during the first fill.

@point2: yes, 1.33 TB platters including the capacity boost from shingling. I'd like to know how large this boost is. Based on traditional 1 TB platters it would be 33%. I think Seagate claimed around 20%, which would match my assumption of 1.1 - 1.2 TB platters without shingling.

And thanks for your other information!

Ah, now I get what you're getting at. Your numbers make sense, and while that number seems a bit low, I do recall Seagate advertising reliable retrieval as a core feature of this product so I'm guessing they were fairly conservative this time round to ensure increased reliability. I remember the early days of PMR and thinking around that time too that the jump from non-PMR to PMR drives seemed too small to be worth it - but again that was just conservatism with early generations of the tech.

Also, in relation to this comment:

Can we conclude form this that the landing zone is not at the beginning of the platters, but further in? At the position where STR drop to ~150 MB/s. Looking at the read STR graph from overclockers.at show this to be the case around 4.6 TB, or "roughly in the middle" of the platters. It also matches the average read speed of 152 MB/s very well. Apparently Seagate choose to optimize average acess times by placing the landing zone in the middle, while sacrificing 40 MB/s write speed. Sounds like a good general purpose trade-off.

I think we now have conclusive evidence the landing zone is at the edge of the platters:

(From: http://hardware.slashdot.org/story/15/03/02/2317227/new-seagate-shingled-hard-drive-teardown)

Share this post


Link to post
Share on other sites

Alright, I've completed one (crude) full benchmark of the disk, doing 8TB of continuous, sustained writes of pseudorandom data (1MB blocks). It seems sustained write behaviour, similar to what you'd get when rebuilding a RAID array with a large stripe size and no other activity is pretty much exactly the same as you'd get on a normal, non SMR drive:

ST8000AS0002.png

Red line = Write speed (MiB/sec), 1 second sample interval

Blue line = Write speed (255-second moving average)

Black line = Excel best fit 2nd order polynomial

X-axis = Position on disk (GiB)

8001563222016 bytes (8.0 TB) copied, 54057.8 s, 148 MB/s

Seems like write speed pretty much exactly matches read speed and Overclockers.at's HDTune result was just... wrong.

I'm running a few other tests with different block sizes to see if that confuses things, but one thing with HDD technology that's gotten worse over the last decade is the speed/capacity ratio... Capacities have increased eight fold while STR has only increased two-fold, making each full benchmark take over 12 hours.

Share this post


Link to post
Share on other sites

Thanks, this looks far more encouraging (and logical) than the RAID rebuild times from the SR review. At least we can conclude that there are sustained transfers which don't make the drive stutter.

MrS

Share this post


Link to post
Share on other sites

We're also trying to get ahold of the new Exblox unit that may handle these drives interestingly as well with its object store.

Share this post


Link to post
Share on other sites

We're also trying to get ahold of the new Exblox unit that may handle these drives interestingly as well with its object store.

Didn't your recent review find the Exablox was god-awful slow even with "normal" 7200RPM drives? Or do you mean they have a new model out that's supposed to be faster?

P.S. I wonder if the RAID rebuild on the SR review was done with other activity on the array? As the performance seen could be expected if the 20GB landing zone gets filled with semi-random writes and the drive has to do garbage collection during the rebuild, and even the HGST rebuild performance was all over the map. But that aside, contrary to the SR review, my findings suggest the drive has no problems at all with sustained sequential writes in isolation.

Share this post


Link to post
Share on other sites

Turns out the drive is pretty terrible for ZFS rebuilds - taking 50 hours for ~2.8TiB, and a similar throughput pattern to that in the SR review. However this is overwriting random data in-place, and ZFS is also very different (and less sequential) than a normal RAID rebuild. I'm going to retest on a freshly zero'ed drive to see if it does make a difference.

Share this post


Link to post
Share on other sites

022 is probably a newer revision of the base model (002). According to the spec sheet it uses 0.1w less poiwer. Some firmware tweaks and firmware level functionality bumps too, particularly for hosts that have SMR aware optimisations.


Now that you mention it, there's quite a of new technical information in the manual for the 022 model. They appear to have published the exact shingling and non-shingled geometries:

1.0 Introduction

These drives provide the following key features:
• Host aware, optimized for SMR performance and capable of ZAC command support
• High instantaneous (burst) data-transfer rates (up to 600MB per second).
• Streaming video optimization - consistent command completion times & ERC support
• Idle3 power mode support
• TGMR recording technology provides the drives with increased areal density.
• State-of-the-art cache and on-the-fly error-correction algorithms.
• Native Command Queuing with command ordering to increase performance in demanding applications.
• Full-track multiple-sector transfer capability without local processor intervention.
• Seagate AcuTrac™ servo technology delivers dependable performance, even with hard drive track widths of only 75 nanometers.
• Seagate SmartAlign™ technology provides a simple, transparent migration to Advanced Format 4K sectors
• Quiet operation.
• Compliant with RoHS requirements in China and Europe.
• SeaTools diagnostic software performs a drive self-test that eliminates unnecessary drive returns.
• Support for S.M.A.R.T. drive monitoring and reporting.
• Supports latching SATA cables and connectors.
• Worldwide Name (WWN) capability uniquely identifies the drive.


1.2 Zone Structure
Archive HDD models use SMR (Shingled Magnetic Recording Technology), physically formatted containing two types of
zones. 64 “Conventional Zones” which are not associated with write pointer, and the media is non-SMR and 29808
Sequential Write preferred Zones which are SMR media. For the sequential write referred zones there is a write pointer to
indicated preferred write location. For the conventional zone writes can occur randomly for any block size. New
commands which report zonal structure, resetting zonal write pointers, as well as managing zonal properties are available
for sequential write preferred zones through ZAC commands.
Archive HDD Conventional Zone Structure
• There are 64 256 MiB Conventional Zones. (ie. Not Shingled)
• The conventional zone is located at the outer diameter and is 16GB.
• Sequential Read and Writes to this zones will perform at similar data rates.
• Random Write commands can be issued in any order without any performance delay.
• Zone designed specifically for random writes data. For example: logs and meta data.
There are 29808 Sequential Write Zones
• Each zone is 2e19 logical blocks in size or 256 MiB each.
• Each zone is a shingled zone.
• To achieve best performance use of ZAC commands is required.
• Re-setting write pointers for each zone is required before reuse.
Optimal number of open sequential write preferred zones
• Advised - the largest number of zones that should be open for best performance, is reported in Identify Device Data
log 0x30 page 0x00h
Optimal number of non-sequentially written sequential write preferred zones
• Advised - the largest number of write preferred zonesthat should be randomly written for best performance, is
reported in identify device data log 0x30 page 0x00h
T-13 standards define the new ZAC commands; REPORT ZONES EXT to query the drive on what zones exist and their
current condition, RESET WRITE POINTER EXT to reset the write pointers, OPEN ZONE EXT, CLOSE ZONE EXT, and FINISH
ZONE EXT to Open, Close, and Finish zones. To achieve optimal performance, an SMR-aware Host driver will need to write
sequentially to all sequential write referred zones.
See the T13 Web Site at http://www.t13.org for ACS-4, T13/BSR INCIT 529 for command details.
  • Like 1

Share this post


Link to post
Share on other sites

Hi guys,

just wanted to confirm, the drive works very slow as a PassThrough on Areca 1680 HBA, sustained write is ~10mb/sec or less. I was trying to copy large video files.

The same speed with reading the files back.

I have removed the drive from HBA and installed it in external SATA dock on Asus Sabertooth X79, and it gives me stable 104 mb/sec write speed from network source. (limited by gigabit connection).

Share this post


Link to post
Share on other sites

I wish I could figure out what I'm doing wrong with these drives. I've got three of them, on two machines (one an old Linux fileserver and one a beefy Windows 7 workstation used for processing RAW photos by the boatload). One will move files onto these drives at a peak of ~60MB/s the other 40MB/s, and this speed will crater soon after transfers start, ending up below 1MB/s for tens of minutes at a time, sometimes just sitting at zero with no indication of activity at all.

Transferring the photo archives, a mix of 25MB RAW files, 1-5MB JPGs and a bunch of tiny settings files, takes something like 3 days for 2TB, if I don't give up before then.

If it was just one computer, or OS, I'd have different suspicions, but as it stands I'm at a loss. Three drives, same problem, and as best I can, I've eliminated the reasons for their poor performance. I have trouble believing your numbers based on my experience. I'd love to find out I'm simply doing it wrong. Any advice would be appreciated.

Share this post


Link to post
Share on other sites

Update: yeah, basically it's the buffer limit. Whatever it is, something around 30-40GB, once it fills everything dies and I have to pause the transfer for 5 minutes while it lays down the new data.

So the drives are great for slowly accumulating new data, but intensely frustrating as replacement backup drives, because that 4TB archive is going to take -ages- to copy over in 30GB chunks.

Share this post


Link to post
Share on other sites

I have purchased this drive (not arrived yet) because I think it will suit my purposes. Can you please confirm if my thinking is correct.

I have 2 x NAS (of 8tb Total of made up of 2tbx4) That I want to back up.. it contains only media (video files) of tv shows and movies I have ripped from my DVDs.

I intend to purchase 2 of these 8th Archive drives to back up each NAS 8tb NAS

I understand the initial 8tb transfer (NAS's are full) will be slow, but once the information is on the disk i can store my backups on an active system in a 2nd location, and the read speeds and retrieval of these files should be sufficient to play back. (hence an active backup)

Occasionally I will reconnect the archive drives to my main system to perform an incremental backup (remove a few files and add a few news ones)

I think this purpose suits the drive, Can anyone please confirm if my thinking is correct.

One last thing about formatting my drives when they arrive.

I am told the drive SMR is self managed and can therefore be used with any operating system so I was going to format the drive in exFAT so that it can be used on both Mac and PC (would this be the right course of action)

Edited by brettwatson

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now