Review: Seagate Archive 8Tb 3.5″ Internal Hard Drive

Hard drive storage space is one thing which always seems to run out regardless of how much you have. Hard drive manufacturers have been working to push the capacity of perpendicular magnetic recording (PMR) drives constantly, but have lately hit a brick wall at capacities of around 4-6Tb. PMR was introduced when drives using longitudinal recording hit their limits at about 320-400Gb, and has proven itself to be relatively reliable with the majority of today’s drives using PMR.

In order to overcome the limits of PMR technology, new technology has to be developed. One approach, taken by HGST in regards to enterprise storage, is to use Helium-filled drives. Helium reduces the friction inside the drive which lowers temperatures, energy usage, improves fly height regulation and allows for extra platters to be incorporated, however, the technology is too expensive for consumer usage and helium is currently in a global shortage which is only likely to get worse over time as it escapes from our atmosphere into space.

Another approach, which is only starting to gain traction, is that of Shingled Magnetic Recording (SMR). This technology is based around PMR, but exploits the differences in the size of the write head versus the read head to improve storage densities.

The Seagate Archive series of drives comes in 5Tb, 6Tb and 8Tb capacities, using 1.33Tb per platter technology. It has a spindle speed of 5900rpm and is qualified for 24×7 operations at 180TB/year workload. It is an Advanced Format 512e drive, and certain models support hardware encryption and instant secure erase. Oriented at large cold-data stores, it features very low power consumption per Gb of data stored, and also has rotational vibration balance for better performance in multi-drive installations. Despite these features, it is also consumer-oriented, being one of the first SMR drives to hit the market, backed by a three year warranty.

Interest in the drive has risen, as they are getting picked up by many computer shops across the country and sold at lower prices down to even AU$309 for 8Tb which seems amazing. But what about the caveats?

Ready to Mingle, with Shingles?

The change to SMR is actually quite a complicated one because SMR drives do not behave like PMR drives. The SMR technique exploits the fact that the write head writes a wider band than is strictly necessary for the read head to recover the data. If we rearrange the non-overlapping tracks in traditional PMR drives so as to overlap partially in writes, leaving just enough of the track for a read, we could squeeze more data into the same area. This looks like “shingles” on a roof, and results in a narrowing of the effective track width.

The issue with such a scheme is that random write access is no longer possible, as the shingled tracks have to be laid down in sequential order to avoid corrupting adjacent tracks.

A storage device that could only be written sequentially would not be easily accommodated by current operating systems, and so, this drive is specially structured and contains its own system to manage the shingle bands.

If you have time, it’s well worth watching this conference talk at Usenix Fast 15 by Abutalib Aghayev and Peter Desnoyers of Northeastern University where an unidentified, but presumed to be Seagate, SMR drive is tested and reverse engineered for its characteristics.

In short, the drive contains a shingle translation layer, which works like the flash translation layer of an SSD to map logical blocks into physical blocks and emulate a block device to the OS with random read/write abilities. The STL consists of a mapping table which keeps track of stale sectors.

In order to perform the miracle of random writes to an SMR drive, the drive contains a persistent cache in the outer diameter tracks, which is basically a “log” of changed sectors. The rest of the drive is arranged so as to have many smaller shingled recording bands separated by guard bands to protect adjacent shingled recording bands.

Random writes are sent to the persistent cache, and when the drive is idle or the persistent cache is full, a “read-modify-write” cycle is effected on the affected shingled recording bands. As a result of the amount of data manipulation required, the drive has a sizable DRAM cache.

As a result, the I/O characteristics of the drive can be quite unpredictable compared to PMR drives, as I/O operations can be cached into DRAM first, then into the persistent cache – and operations to the persistent cache can be interrupted by band-updates causing the drive to “stall” for ~0.6s. Therefore, this drive is not really suited for write heavy environments and should only be used where writes are few, mostly sequential rather than random and interleaved with lots of idle time for the shingled bands to be updated by the drive.

Also, because the drive essentially has its own “brains”, the drive can remain active even when idle and generate noise, heat and power consumption. Use of such a drive in an external enclosure might not be recommended unless you guarantee safe ejection and spin-down of the drive, as an unexpected power-down could in some cases result in possible data corruption. It is also important that the bridge chip be able to handle occasional delayed I/O responses too.

Unpacking the Drive

20150421-1913-4899I purchased two of the drives over a month ago at AU$379, which is a little much. It was one of the first batches to arrive, however, so I thought it deserved the premium. Because the drives were a little special, I decided to spend a whole month testing them before I delivered my verdict.

The drives were shipped to me from Synnex, and came wrapped in a thin bubble wrap bag, and thrown into a box. Seagate won’t accept drives packaged so poorly for RMA, so why are they shipping them out like that?

As usual, the drive comes in an anti-static bag.

20150421-1913-4900

The drive itself comes with a three year warranty, which is more than the two years you get with the regular Desktop series.

20150421-1914-4902

Eight terabytes. How amazing. All within the same 3.5″ standard from factor. This unit was dated Week 3 of 2015, manufactured in Thailand with firmware AR13. This unit is the ST8000AS0002, which does not feature on-drive encryption. It claims to use 0.55A on the 12V rail and 0.35A on the 5V rail, which is fairly economical and “green”.

20150421-1914-4901

The drive has a very “solid” boxy shape, with the “tub” extending right out to the absolute extremes of the 3.5″ form factor. This is because the 8Tb drive consists of six platters, which is pretty much the absolute limit for air-filled drives. As a result of the large size of the platter chamber, there is no mid-mount hole as with regular drives. The PCB has also been specially designed to fit into a small milled-out recess in the side, held in by small Philips screws.

20150421-1916-490820150421-1915-4906

The quirkiness with the drive mounting holes extend to the sides as well, where the mid-mounting hole does not exist. This doesn’t pose a problem for most mounting designs, although you might want to check just in case.

20150421-1914-4904

The front of the drive has a serial number label, which will make guys like Backblaze happy, so they can identify faulty drives quickly.

20150421-1915-4907

At the rear, nothing special, just the normal SATA connectors along with the diagnostic pins.

20150421-1916-4909

The front edge of the drive seemed to have an interesting ‘stepped’ area to the tub – I’m not sure what this is for, although it did get my curiosity.

Digging a Little Deeper

So I have a new and expensive drive, and I’m curious, what powers it? So I decide to dismantle the PCB (a warranty voiding procedure) and take a peek. We’ve already established that there’s nothing on the outward facing side of the PCB, so everything must be hidden underneath.

20150421-2004-4911

Removing the Philips screws allows the board to come free – two main chips have a thermal pad connection to the chassis for cooling. Removing the blue foam padding reveals the rest of the board – not much more.

20150421-2006-4912

It seems there are two silver rotational sensors at the extreme edges of the board, and an EEPROM for firmware.

20150421-2006-4913

The spindle motor control is done by a STMicroelectronics Smooth controller, code-named Dillon, custom designed for Seagate.

20150421-2006-4914

The hard drive controller is an LSI TTB71901VB chip also made for Seagate. This chip appears special because of the double row of staggered pins, which is not common. Next to it is a Samsung K4B1G1646G-BCH9 128MiB DDR3 DRAM as the cache buffer.

Testing It Out

The drives were stressed for a whole month as part of my tests, because I had some reservations about the SMR technology. They were installed in a new HP Microserver N54L, although it only has SATA II ports. This is unlikely to limit the performance of the drives in any way however.

Suffice it to say, no SMART errors were thrown throughout a month of testing on both units. Here is a sample of the SMART attributes reported by one of the units:

seagate8tb-smart

HDTune Pro

Of course, the speed of the drive would be of a prime concern to users, so I performed sequential write and read tests using HDTune. Two drives were used, with two write benchmarks done back-to-back, followed by one read benchmark.

Drive #1

23-April-2015_01-06 23-April-2015_16-29 24-April-2015_08-37

The two write benchmarks are virtually identical, with an average write rate of 145.9MB/s which is quite impressive. The read benchmark mirrors the write benchmark with an average of 146.7MB/s. The impact of the STL can be seen in the access time write benchmark, where unusually low figures of <1ms are reported as the drive’s STL acknowledges writes before they actually happen.

Drive #2

24-April-2015_23-51 25-April-2015_15-12 26-April-2015_06-37

The second drive scores very similar results, although the patterning in the transfer rate is subtly different because of drive-to-drive variance in the sector layout/density and the head to media match.

Comparison

For comparison, two new 4Tb PMR Seagate Desktop drives also had write and read benchmarks run on it using the same system.

26-April-2015_17-23 27-April-2015_01-46 27-April-2015_15-25 27-April-2015_23-40

The two PMR drives had significant variances in throughput in the outer diameter, likely because the second drive had poorer media to head match. The read and write speed averaged {137.5MB/s, 134.3MB/s} and {137.2MB/s, 134MB/s} respectively. The average transfer rate of the 8Tb drives were slightly faster at about 145MB/s which is a pleasing sign, although considering the areal density difference, a greater speed increase may have been expected.

For more comparisons, please see this article where WD and Seagate 4TB PMR drives are tested.

ATTO Benchmark

Looking at the block I/O performance, we find some very strange results with ATTO which is probably a result of the STL itself. Traditional disk benchmarks don’t seem to play that well with the SMR drives because of the persistent cache and DRAM cache interfering with the test.

Drive #1

atto-8tba-12

Drive #2

atto-8tba-11

Comparison

For comparison, please see this article where WD and Seagate 4TB PMR drives are tested. The PMR drives provide much more consistent and predictable I/O performance over the block sizes – the exaggerated write performance is almost certainly the result of the large DRAM cache, persistent cache on the outer tracks and STL acknowledging writes before they have actually hit the disk. This is not helped by the small test-file size. The inconsistency in the results may reflect the differing internal states of the two SMR drives at the point of testing.

CrystalDiskMark

Drive #1

cdm-8tb-a

Drive #2

cdm-8tb-b

Comparison

For comparison, please see this article where WD and Seagate 4TB PMR drives are tested. The drives seem to turn out pretty good numbers overall, although quite inconsistent between the two drives, but again, because of the small test file size and the design of the STL, the drive puts out artificially good numbers. As a result, regular benchmarks may not be so applicable to self-managed SMR drives.

H2testW

H2testW is normally used to test flash drives for data integrity, however, we can also use it to test hard drives. Unfortunately, because it doesn’t seem to be able to compute the size of the MFT after the test files are placed on the drive, the Write+Verify test fails with an error that the drive filled up quicker than expected (as the MFT grew due to the number of files written by the test).

8tb-h2tw-error

However, this is no problem as the data itself can be verified.

Drive #1 & #2

The average verify rate is about 128MB/s which is not too bad, and could be CPU-related too.

8tbb-h2tw-ok 8tba-h2tw-ok

More importantly, the test finished without any detected data corruption. This is especially difficult, as these drives are specified for one unrecoverable read error in 10^14 bits read. This translates to one read error in 113.69TiB/125TB read. This statistic has not really moved in all the years, so as drives get bigger, read errors might be encountered more frequently. Using that specification, you might expect to see one read error after reading the drive 15.63 times.

Comparison

For comparison, please see this article where WD and Seagate 4TB PMR drives are tested. On the whole, transfer rates in the test seem fairly similar for this test.

Other Notes

The drive performed solidly during a month of testing, and didn’t have any big heat or vibration issues. The one annoyance is a tendency for the drive to be “clicky/wheezy” on seeking and seeking a lot because of the SMR band rewrites, thus sometimes it sounds “busy” even when it shouldn’t be. It’s not that loud, and I could easily sleep with these running in my room, but you probably wouldn’t want it in an HTPC.

Because of the SMR difference, you should also be careful in the way you use the drive. It’s absolutely fine for reading, but writing performance can take a serious hit under heavy random write access and is less consistent compared to PMR drives. It’s best suited for a single-user mostly-sequential large-file write scenario (where write rates of 69MB/s in small blocks, and up to 188MB/s in large blocks are seen on the outer tracks). It isn’t ideal for multi-user NAS scenarios due to the hit that random writes cause (multiple shingle band updates in each write). However, if you load up a fixed set of data and mount the partition read-only, I see no reason why you can’t use it for multi-user read-serving, although as a “green” style drive, you do expect it to be somewhat sluggish.

It’s also not best suited for real-time applications, such as video recording, surveillance monitoring or RAID systems. I also question its applicability to external enclosures, as some chipsets may not be ready for a drive of 8Tb, and they will have their own sector translation rules as well. Aside from that, unsafe power-down and unexpected ejection could have data loss ramifications with an extra level of caching in this drive, and some cases may spin down the drive on idle, reducing the amount of available idle time it can use to reconcile its persistent cache into the shingled recording bands.

Despite these caveats, after a thorough consideration of how most media tank drives operate, the Archive seems ideal for large scale media storage. The data rarely changes (if ever) and write operations are few, and mostly large sequential writes in nature. The media drives generally have lots of idle time and read-mostly workload. This is the same of a back-up drive as well. These are ideal applications for a drive of this sort, and SMR technology has no real drawbacks when it comes to this.

Conclusion

The Seagate Archive series of SMR drives offers a lot of storage for an affordable price. As these drives are SMR, their performance characteristics do differ from that of regular PMR drives.

The drives are ideally suited for large data stores, where read access is most common, and lower levels of write access are expected. It performs best when writes are large sequential writes, as opposed to scattered small random writes, and needs some level of idle time to ensure the bands are updated. In most home media applications, this is ideal as most media servers are idle and reading is the predominant operation.

That being said, even substantial rates of sustained writes can be handled by the drive, benchmarks showing anywhere from 69MB/s on the outer track in small blocks, up to 188MB/s in large blocks, so the drive isn’t much of a slouch at all. The read performance closely matches the write performance, and the disadvantages of SMR don’t show in most of the regular benchmarks, partly because the shingled bands are relatively small and fast to rewrite.

Of course, due to the I/O inconsistencies and the specific operations of the STL, it is not recommended this drive be used for real-time I/O (say video capture) or RAID applications. I wouldn’t recommend external drive applications either, partly due to sector translation issues possibly arising, and unexpected powerdowns may cause data corruption. It also isn’t an ideal drive if noise is your concern, as this drive has more seek noise than expected due to the six platter construction and the need to seek back and forth between persistent cache and shingled recording band areas. Heat was not an issue, however.

As for long term reliability, only time can tell, and there is always a risk with using “first generation” hardware. That being said, I did not experience any issues at all during my month of testing, and it is based upon a modified form of PMR which we have “mastered” over the years. The three year warranty and enterprise-and-consumer targeting of the drive inspires some level of confidence, as the regular desktop drives get just two-years.

I suppose this does make the Seagate Archive an attractive drive for those looking to back-up large amounts of data or run a large home-media store, at least, until the next technology comes about – which might be Heat-assisted Magnetic Recording. I’ve got a third drive on the way …

Posted in Computing | Tagged , , , , , | 5 Comments

Teardown & Test: Nelson Cougar 60 Electronic Downlight Transformer

You never quite go to Bunnings without leaving with quite a few items. While walking around the aisles, the Nelson Cougar 60 caught my eye.

20150524-1220-5266 20150524-1220-5267

This is an electronic transformer designed for use with halogen downlights up to 60 watts. The big surprise, at least for me, is that a single transformer is being sold for AU$7.86 at retail, with ten packs at AU$61.98. You can even get a version with a plug and lead for use with GPOs for AU$10.33. There are also Osram Redback 40VA units available for a little more, and higher-end Nelson “tradesman” models. I remember when electronic transformers were much more expensive than this, which was why I never really ever played with these units. As a result, I decided to throw some money at it, despite not really needing it, just to see what ~$8 gets you.

The Item

The unit itself is very small, about palm size. I’ve never met a transformer lighter or smaller. That’s what technological progress brings you!

20150524-1221-5268

It has a two tone curved plastic motif, which is pretty trendy for something that’s virtually never seen and should never be heard. The model and specifications are printed on the top, although slightly blurred. It has a high power factor rating of >0.99. It claims a level III efficiency rating (>=84% under load with <=0.75W idle), which isn’t as efficient as it could be. The output is unusual, as it claims to be 12.5v (a little higher than 12v) which is the opposite of most transformers, which put out a little less than 12v to ensure longer filament life. The unit is dimming compatible, with trailing edge dimmers.

20150524-1221-5269

The secondary contacts are left open at the end to the outside, whereas the primary contacts are enclosed by a plastic screw-down cover, which also has side “click” latching. The cover needs to be pried off to install the cable initially.

20150524-1222-5271

The opposite side is marked 15W04 230, which may be a specific model number.

20150524-1221-5270

A flat base is provided for mounting against flat surfaces. No ribbing or elevation is provided, which implies that the transformer doesn’t put out much heat. Two screws are used to hold it down – one fixed end, and the other as a slot to accommodate slight drilling errors. The design is a simple plastic casing with no screws used in the construction.

Teardown

The unit comes apart once you remove the mains cable cover, and then pry around the white to blue plastic seam, where four “clips” are used to hold the cover down. Internally, the PCB is wrapped by a plastic insulating sheet.

20150524-1227-5272

The PCB is a single layer paper type PCB, with green solder resist and limited silk-screening on the underside. The solder joints on the underside look very consistently made. A clear demarcation between primary and secondary is visible. The PCB is marked ET60W-5.

20150524-1227-5273

From the top, we can see there are no heatsinks and no ICs throughout the design. There are two main transistors, and a pair of transformers (one primary/secondary, and a feedback) which are responsible for the operation of the device. To support this, there are a handful of diodes, resistors, capacitors and a single transistor.

20150524-1228-5275

The input appears to be protected by a fuse wrapped in heatshrink, and an inductor provides some RF filtering.

20150524-1227-5274 20150524-1228-5276

Components near the edge of the PCB are covered with a woven tube for fire protection.

20150524-1228-5277

The output has a very unusual winding. For double-insulation isolation reasons, the inner winding is separated from the outer secondary by a plastic shell, and exceeding requirements, is the use of insulated wire for the outer winding. This loose coupling may be part of the reason for a slightly lower efficiency than expected.

Output Waveform

When the transformer is powered with no load, it makes an audible buzz. Checking the output shows a spiky output which follows the mains envelope, much like the Osram Redback transformer, but with a limited peak voltage of about 14v and an AC RMS value of 3.48v.

unloaded-output-1

Unloaded, the individual spikes seem to cycle at a rate of 5.5khz, and shows a characteristic “ringing” style waveform that you might find from a ringing choke converter, which gives you an idea of the simple regulation method used.

unloaded-output-2

Loaded with a 50W (nominal) halogen downlight globe, the transformer performed superbly, producing an average of 12V RMS AC over the globe – perfect.

50w-halo-load-0

Of course, this means a peak voltage of about 18v, but this is of little significance as the filaments of the globe are more sensitive to the RMS value which is accurate – although less than the claimed 12.5v. With the correct load, the converter “rings” quicker at 51.72khz. The waveform resembles more of a square wave.

50w-halo-load

Reducing the load with a 10W Philips LED replacement, the transformer still ran without audible noise, and the LED globe seems to work happily. The output from the transformer seems to contain taller spikes with ~25v peaks. This is why marrying electronic transformers with LED replacements can sometimes result in unexpected failures.

10w-LED-load

The waveform oscillates even quicker at 66.86khz, with two components to it. It is nice to see that it works though.

10w-LED-load-2

Conclusion

This electronic transformer seems to behave similarly to the Osram Redback 60VA transformer I tore down before. For the price and weight, electronic transformers are a marvel. There isn’t even a single IC in the design, which reminds me of the ballast designs of CFLs. Their output, however, is more unpredictable from a waveform and frequency standpoint, so it isn’t that useful for repurposing. That being said, for ~$8, it definitely works, and only buzzes when unloaded.

Posted in Electronics, Lighting | Tagged , , , | Leave a comment

Review & Test: Philips LEDbulb B22 14w 1400lm LED Globe

Sometimes you need to solve your own problems around the house, and the best place to visit is a hardware store. Just last week, I took a trip to Bunnings, and I couldn’t resist surveying their array of LED light globes. While consumers are still at a cross-roads when it comes to the CFL versus LED debate, LED technology and prices have continued to improve.

The Product

20150524-1217-5259Since I purchased my first set of Philips LED globes from Woolworths, it seems that Philips’ product line-up has improved. The 13W globe offered 1055 lumens, resulting in an efficacy of 81 lumens per watt, the new 14W LEDbulb offers 1400 lumens for an efficacy of 100 lumens per watt.Of course, these numbers continue to improve with each generation, as the efficiency of LEDs continue to improve. This is great news for those looking for a little more light than the original LED globes could offer.

20150524-1217-5260

 

Of course, Bunnings has other LED globes on offer as well, many of them in smaller power ratings. The other LEDs are mainly from Osram which top out at 10.5W. As someone who is looking for higher powered globes, this wasn’t of much interest.

I purchased two samples of the bayonet cap (B22) mount globes, at AU$18.99 a piece, in order to cover the rest of the house as we have a mixture of fixtures. This is a bit pricey compared to CFLs, but still not out of the reach of individuals.

As seems to be common with LED globes, the choice of colour temperature is limited. Bunnings only stocks Warm White 3000K globes at 14W in both BC and ES, but also stocks a Cool Daylight 6500K 14W globe in ES only.

The claims made on the packaging haven’t changed much, with an 80% claimed energy saving and up to 15 year lifetime. It also claims light “equivalency” to about 90w.

20150524-1218-5261 20150524-1218-5262

20150524-1218-5263The packaging has only suffered subtle changes, with the globe having a claimed lifetime of 15,000 hours, which is a little bit on the low side when it comes to LEDs. This may be due to the lack of external-facing heatsink, which is enclosed within a plastic shell, and due to their conservatism. The Warm White 3000K is slightly less warm compared to the 2800K offered by the IKEA globes. There is a change which claims you should use in >4″ diameter downlight, as opposed to 3″ previously. The globe remains non-dimmable, and is Made in China.

20150524-1218-5264

From the outside, the globe appears the same as the 13w version, with an elongated enlarged shape resembling that of an old GLS globe. Due to the larger size, it may not fit in all fittings. When picked up, the globe feels a little on the lighter side compared to the IKEA 13w globes, which may indicate a slightly smaller metal content, implying a smaller heatsink.

20150524-1219-5265

As expected, the branding and specifications are printed around the neck in grey. It is claimed to operate on a 220-240v range at 50/60Hz, drawing 80mA. Due to the shape of the globe, it seems that the globe is best used on light fixtures where the globe is mounted with the base up, so the diffuser points right at the ground.

As our dwelling uses a vast array of oyster alabaster fixtures, I had to forego the alabaster cover altogether because the globe would not physically fit otherwise. As it turns out, because of the sideways mounting, the lighting in one half of the room was much more than the other half. As a result, I handmade a right-angle adapter based around a short piece of flex, a suspension bayonet cap holder and a bayonet cap adapter, like a very short suspension kit.

20150524-1231-5286

This allowed for a much better light utilization, although with a little glare due to the directness of the light despite the integrated diffusing dome. The light output was relatively impressive – I didn’t note any issues with flicker or brightness, and the colour temperature was as expected. However, the colour rendering index was not stated, and I didn’t have the equipment to measure it, but a difference in colour rendering was noted which made the light appear a little stark and colder, so I’d probably guess a CRI of >82.

Electrical Performance

The performance of the previous 13w globe was relatively middle-of-the-road, showing a loose regulation where the power consumption is strongly dependent on input voltage. The power factor was about 0.8, which is higher than a poor quality CFL, but hardly as good as it could be.

Philips-13w-LED-ES

The new globe improves upon this performance somewhat, with a higher power factor of about 0.9 across a broad voltage range. The regulation is still relatively lax, with the power consumption still strongly related to the input voltage. The claimed 14W is a bit of an under-rating with the consumption about 14.1W at 220v, 14.4W at 230v and 14.8W at 240v. As a result, it’s probably more of a 15W globe.

Philips-LEDbulb-14w-PowerAnalysis

From the voltage range being “shrunk”, it seems likely this globe consists of more LEDs connected in a series string, rather than less LEDs connected in a series-parallel arrangement. It also seems likely that the current driver design has changed.

As I don’t have the equipment to measure absolute light output, CRI and colour temperature, these parameters cannot be determined.

Conclusion

While LED globes at retail are being sold at high prices that make their prices difficult to justify based on their claimed lifetimes, it’s always nice to see that they are being offered as an alternative. LEDs reach full brightness instantly, and suffer very little from power on-off cycles which can be very desirable, and the lack of mercury is a positive for the environment.

It seems that Philips has improved upon their former design with a higher power factor current driver, and a much improved efficacy (+23%) at a similar price-point. The current driver does a lax job of regulating power consumption over different line voltage conditions, but doesn’t seem to have any flicker issues. The short claimed lifetime may be down to conservative estimates, or the use of an “enclosed” heatsink with low metal mass, which makes purchasing LEDs for their longer lifetime an unclear proposition.

Posted in Electronics, Lighting | Tagged , , , , , | 4 Comments