News:

Willkommen im Notebookcheck.com Forum! Hier können sie über alle unsere Artikel und allgemein über Notebook relevante Dinge disuktieren. Viel Spass!

Main Menu

Post reply

The message has the following error or errors that must be corrected before continuing:
Warning: this topic has not been posted in for at least 120 days.
Unless you're sure you want to reply, please consider starting a new topic.
Other options
Verification:
Please leave this box empty:

Shortcuts: ALT+S post or ALT+P preview

Topic summary

Posted by Lsi
 - May 04, 2020, 05:43:01
In case the meaning of my CDM results isn't obvious enough, the Seagates maintain normal read performance when their SMR-compensating write cache is disabled, and still scale up properly under QD32 random 4k writes.

Everything tanks on the WD, and random writes slow to a trickle at 40 IOPS regardless of light (QD1) or heavy (QD32) load conditions.  Its random QD32 4k reads appear to improve, but that is only because the aggressive random QD1 4k cache-enabled writes leaves the drive doing background "writes catch up" during the 4k QD32 read test.
Posted by Lsi
 - May 04, 2020, 05:33:43
If SMR is well designed firmware / cache wise, it allows large capacity products to hit the market faster and with equivalent or slightly better performance under light loads--since these products usually have 128MB+ cache and (hopefully) media caching as well.

The easy way to determine if a SMR product will fall flat under heavier loads is to run a normal (1GB) Crystal Diskmark test--but with the write cache disabled.  Modern Seagate 2.5" SMR products I've tested (which have been on the market for years) still continue at a reasonable speed, but WD 2.5" products such as their WD10SPZX run horribly.  Even its sequential read performance drops severely, so its cache algorithms are very badly designed for handling write saturation.

I haven't tested a newer gen WD Red, but after running these tests I'm not surprised at all.  Furthermore, using SMR on a drive in NAS/RAID is simply a bad idea in general because of the expectation of sustained write load during a rebuild scenario.  Until you are absolutely sure that your cache algorithms (even with write caching disabled) can deliver reasonable QoS / latency / throughput during that type of load, this type of product should not be marketed.

I'm attaching the WD10SPZX (5.4k rpm) results with its write cache on and then off, and then two SMR Seagates (ST1000LM035-5.4k and ST500LM034-7.2k) with write caching disabled for perspective:

WD10SPZX firmware 201a02, write cache on:
-----------------------------------------------------------------------
CrystalDiskMark 3.0.4 x64 (C) 2007-2015 hiyohiyo
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read :   128.392 MB/s
          Sequential Write :   124.431 MB/s
         Random Read 512KB :    43.944 MB/s
        Random Write 512KB :   109.451 MB/s
    Random Read 4KB (QD=1) :     0.504 MB/s [   123.0 IOPS]
   Random Write 4KB (QD=1) :     9.036 MB/s [  2206.0 IOPS]
   Random Read 4KB (QD=32) :     0.611 MB/s [   149.2 IOPS]
  Random Write 4KB (QD=32) :     1.272 MB/s [   310.5 IOPS]

  Test : 1000 MB [D: 0.0% (0.1/931.5 GB)] (x5)
  Date : 2020/05/03 14:54:46
    OS : Windows 10 Professional [10.0 Build 18363] (x64)
 
WD, write cache off:
-----------------------------------------------------------------------
CrystalDiskMark 3.0.4 x64 (C) 2007-2015 hiyohiyo
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read :    48.871 MB/s
          Sequential Write :    11.078 MB/s
         Random Read 512KB :    17.899 MB/s
        Random Write 512KB :    20.634 MB/s
    Random Read 4KB (QD=1) :     0.457 MB/s [   111.5 IOPS]
   Random Write 4KB (QD=1) :     0.161 MB/s [    39.4 IOPS]
   Random Read 4KB (QD=32) :     0.966 MB/s [   235.8 IOPS]
  Random Write 4KB (QD=32) :     0.162 MB/s [    39.5 IOPS]

  Test : 1000 MB [D: 0.0% (0.1/931.5 GB)] (x5)
  Date : 2020/05/03 14:34:40
    OS : Windows 10 Professional [10.0 Build 18363] (x64)
 
ST1000LM035 firmware RSM8, wc disabled:
-----------------------------------------------------------------------
CrystalDiskMark 3.0.4 x64 (C) 2007-2015 hiyohiyo
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read :   146.062 MB/s
          Sequential Write :    16.305 MB/s
         Random Read 512KB :    47.639 MB/s
        Random Write 512KB :     9.714 MB/s
    Random Read 4KB (QD=1) :     0.574 MB/s [   140.1 IOPS]
   Random Write 4KB (QD=1) :     0.212 MB/s [    51.8 IOPS]
   Random Read 4KB (QD=32) :     1.328 MB/s [   324.3 IOPS]
  Random Write 4KB (QD=32) :     1.184 MB/s [   289.1 IOPS]

  Test : 1000 MB [D: 0.0% (0.1/931.5 GB)] (x5)
  Date : 2020/05/03 18:19:45
    OS : Windows 10 Professional [10.0 Build 18363] (x64)
 
ST500LM034 firmware SDM2, wc disabled:
-----------------------------------------------------------------------
CrystalDiskMark 3.0.4 x64 (C) 2007-2015 hiyohiyo
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read :   172.861 MB/s
          Sequential Write :    22.112 MB/s
         Random Read 512KB :    56.960 MB/s
        Random Write 512KB :    12.890 MB/s
    Random Read 4KB (QD=1) :     0.682 MB/s [   166.5 IOPS]
   Random Write 4KB (QD=1) :     0.273 MB/s [    66.6 IOPS]
   Random Read 4KB (QD=32) :     1.453 MB/s [   354.7 IOPS]
  Random Write 4KB (QD=32) :     1.242 MB/s [   303.2 IOPS]

  Test : 1000 MB [D: 0.0% (0.1/465.8 GB)] (x5)
  Date : 2020/05/03 15:27:18
    OS : Windows 10 Professional [10.0 Build 18363] (x64)
 

A PCMark benchmark on the ST1000LM035 with its write cache disabled dropped to ~50% of its write cache enabled score--which is certainly acceptable performance for SMR in its "worst case" write performance state.  With its write cache enabled, its PCMark score was slightly higher than an older generation WD MZ500S 7.2k 2.5" CMR drive with 32MB cache.

I would describe Seagate's modern SMR implementation as "done right" for average client use / desktop scenarios, but I still wouldn't want these compromises in a NAS product.
Posted by S.Yu
 - April 21, 2020, 15:01:41
Quote from: not_anton on April 20, 2020, 09:10:54
Bought a 2TB Toshiba laptop drive because I needed more space in an old laptop. Did not have a word about SMR in its description.

Windows 10 installation took 6 hours and failed to boot... A waste of money!
I don't think a drive as small as 2TB even needs SMR, perhaps 2.5" >4TB and certainly >8TB 3.5" is where you need to be careful. Whatever your issue is it's not SMR.
Posted by S.Yu
 - April 21, 2020, 14:59:53
Quote from: Craig Ward on April 20, 2020, 03:06:47
@S.Yu - That's part of the problem.

We're not focusing as heavily on Toshiba because they're just using SMR in some value products without labelling them as such.

However, Seagate has started using SMR in several Barracuda models that previously used faster conventional recording methods - such as 2 TB, 4 TB, and 8 TB drives with the regular STXXXXDM00X model numbers e.g. ST2000DM008.

Similarly, Western Digital has been selling WD Red 2 TB, 4 TB, and 6 TB drives using SMR with model numbers ending in 'FAX' vs the older 'FRX' models. The Red drives were all listed as designed to work together in a NAS, but these new drives have been causing rebuild issues when being added to existing setups.

So in the case of Seagate and Western Digital, unless people are very careful at comparing model numbers, they would assume that the Barracuda or the Red they were about to buy in 2020 was the same as the review they had just seen from 2018/2019 about the same capacity Barracuda or Red.

Further to this, because of the use of a cache platter, benchmarks that only write a few dozen GB of data or less (e.g. CrystalDiskMark) would show the same full write performance as the older drives. You would need to write a larger amount of data than the cache to see the difference in speed. This probably wouldn't matter for the majority of users out there, but in the case of the NAS drives, it can impact heavily on performance for activities that involve large amounts of data (e.g. adding a blank drive to a pool, creating a full backup from another device, or moving large folders/projects onto the NAS).
Oh, I see, but at least when I do my own testing I always write >100GB to be able to observe any cache behavior.
Posted by not_anton
 - April 20, 2020, 09:10:54
Bought a 2TB Toshiba laptop drive because I needed more space in an old laptop. Did not have a word about SMR in its description.

Windows 10 installation took 6 hours and failed to boot... A waste of money!
Posted by Craig Ward
 - April 20, 2020, 03:06:47
@S.Yu - That's part of the problem.

We're not focusing as heavily on Toshiba because they're just using SMR in some value products without labelling them as such.

However, Seagate has started using SMR in several Barracuda models that previously used faster conventional recording methods - such as 2 TB, 4 TB, and 8 TB drives with the regular STXXXXDM00X model numbers e.g. ST2000DM008.

Similarly, Western Digital has been selling WD Red 2 TB, 4 TB, and 6 TB drives using SMR with model numbers ending in 'FAX' vs the older 'FRX' models. The Red drives were all listed as designed to work together in a NAS, but these new drives have been causing rebuild issues when being added to existing setups.

So in the case of Seagate and Western Digital, unless people are very careful at comparing model numbers, they would assume that the Barracuda or the Red they were about to buy in 2020 was the same as the review they had just seen from 2018/2019 about the same capacity Barracuda or Red.

Further to this, because of the use of a cache platter, benchmarks that only write a few dozen GB of data or less (e.g. CrystalDiskMark) would show the same full write performance as the older drives. You would need to write a larger amount of data than the cache to see the difference in speed. This probably wouldn't matter for the majority of users out there, but in the case of the NAS drives, it can impact heavily on performance for activities that involve large amounts of data (e.g. adding a blank drive to a pool, creating a full backup from another device, or moving large folders/projects onto the NAS).
Posted by S.Yu
 - April 20, 2020, 00:18:43
I always look up the RW of a drive before purchase, so this trick would never work on me, and if they mixed slow drives in with fast drives, I would definitely return until I get a fast one.
Posted by Apu
 - April 18, 2020, 18:27:07
Shady. Specs or not, this is something the customers would like to know about.
Posted by Jason Freeman
 - April 18, 2020, 17:54:36
i would still buy it but it would it would have a cheaper  price or 3tb hard drive as advertised 
   
Posted by Redaktion
 - April 18, 2020, 13:15:01
Western Digital, Seagate, and Toshiba have been in the news recently for using SMR technology in drives without advertising them as such. In the case of Western Digital and Seagate, it consisted of substituting SMR drives into product lines that didn't previously use the technology, impacting performance and causing issues in certain use cases.

https://www.notebookcheck.net/Western-Digital-Seagate-and-Toshiba-have-been-using-slower-technology-in-some-recent-hard-drives-without-telling-customers.461822.0.html