Patrick has an interesting video on HDD endurance, prompted by the relatively low endurance of the WD Red Pro drives:
watch
It is an interesting twist to look at the poor endurance of a spinner when compared to that of an SSD, especially with a preponderance of reads vs writes. Clearly he has a point and it is not the first time an issue has arisen with the rapid growth in HDD capacity revealing the underlying architecture limitations of the average SATA hard drive.
But the endurance rating limitation of HDDs being governed by the total of writes + reads is a serious issue (especially for the WD Pro and Plus) and he mentioned the impact of read-intensive background tasks such as ZFS scrubs, RAID amplification, management etc before we even get to the business of actually reading and writing data for client use.
Although not mentioned specifically, this looks equally applicable to DSM and BTRFS scrubs - reading all the data on a drive and comparing it with the metadata checksums. It is recommended to complete BTRFS scrubs every month and those read tasks will eat through the yearly endurance rating. As a simplistic example, BTRFS scrubs on a 90% full WD Plus and WD Pro in a single year:
14 TB WD Red Plus, endurance limit 'up to' 180 TB/year @ 90 % full
= 12.6 TB write + (12.6 TB read x 12 months)
= 163.8 TB
~ 91% of endurance used
20 TB WD Red Pro, endurance limit 'up to' 300 TB/year @ 90 % full
= 18 TB write + (18 TB read x 12 months)
= 234 TB
~ 78% of endurance used
As the above figures do not include any RAID, DSM or task overheads they are artificially low but it is clear to see that even a drive you write to once and do nothing more than data integrity checks will burn through the overwhelming majority of its 'up to' endurance in any given year.
In the video Patrick is his usual effervescent self but he does seem to have a very good point to make - it is remarkably easy to burn through the yearly endurance ratings. In the case of the Red Plus example above a drive filled just once does not have sufficient endurance remaining to do standard BTRFS scrubs, let alone all the other routine drive and data management tasks required of it. The Red Pro fairs little better as in reality as things like actually using the data plus more intensive RAID tasks (RAID expansion anyone?) will blow through the limit.
For those spinning WD Reds, should the recommended monthly BTRFS scrub be dispensed with? Is a scrub every 3 months enough to provide a meaningful balance of integrity and stay under the endurance limit? Is a 6-monthly scrub the maximum practical limit to preserve a drive for real-world use? Are drives manufactured with higher limits (eg 550 TB/year) also at risk of hitting their endurance limits?
I have always likened these massive SATA HDDs as a large house that you can only fill and empty via the letterbox; now it looks like the postman has serious limitations too.
️
watch
It is an interesting twist to look at the poor endurance of a spinner when compared to that of an SSD, especially with a preponderance of reads vs writes. Clearly he has a point and it is not the first time an issue has arisen with the rapid growth in HDD capacity revealing the underlying architecture limitations of the average SATA hard drive.
But the endurance rating limitation of HDDs being governed by the total of writes + reads is a serious issue (especially for the WD Pro and Plus) and he mentioned the impact of read-intensive background tasks such as ZFS scrubs, RAID amplification, management etc before we even get to the business of actually reading and writing data for client use.
Although not mentioned specifically, this looks equally applicable to DSM and BTRFS scrubs - reading all the data on a drive and comparing it with the metadata checksums. It is recommended to complete BTRFS scrubs every month and those read tasks will eat through the yearly endurance rating. As a simplistic example, BTRFS scrubs on a 90% full WD Plus and WD Pro in a single year:
14 TB WD Red Plus, endurance limit 'up to' 180 TB/year @ 90 % full
= 12.6 TB write + (12.6 TB read x 12 months)
= 163.8 TB
~ 91% of endurance used
20 TB WD Red Pro, endurance limit 'up to' 300 TB/year @ 90 % full
= 18 TB write + (18 TB read x 12 months)
= 234 TB
~ 78% of endurance used
As the above figures do not include any RAID, DSM or task overheads they are artificially low but it is clear to see that even a drive you write to once and do nothing more than data integrity checks will burn through the overwhelming majority of its 'up to' endurance in any given year.
In the video Patrick is his usual effervescent self but he does seem to have a very good point to make - it is remarkably easy to burn through the yearly endurance ratings. In the case of the Red Plus example above a drive filled just once does not have sufficient endurance remaining to do standard BTRFS scrubs, let alone all the other routine drive and data management tasks required of it. The Red Pro fairs little better as in reality as things like actually using the data plus more intensive RAID tasks (RAID expansion anyone?) will blow through the limit.
For those spinning WD Reds, should the recommended monthly BTRFS scrub be dispensed with? Is a scrub every 3 months enough to provide a meaningful balance of integrity and stay under the endurance limit? Is a 6-monthly scrub the maximum practical limit to preserve a drive for real-world use? Are drives manufactured with higher limits (eg 550 TB/year) also at risk of hitting their endurance limits?
I have always likened these massive SATA HDDs as a large house that you can only fill and empty via the letterbox; now it looks like the postman has serious limitations too.
️