Brian

Seagate Ships 1MM SMR Drives Discussion

13 posts in this topic

After reading the tech article it sounds to me that these drives will be best for infrequently modified large data storage. I can't imagine how long defragging might take.

This tech might do best merged with a large solid state cache.

Share this post


Link to post
Share on other sites

We're thinking the same way...Seagate hasn't been very open about where these drives are going and what they're doing unfortunately. But I suspect as you do that it's a very specific workload where these drives can succeed.

Share this post


Link to post
Share on other sites

Perfect for the cold storage solutions Facebook is asking around for a while.

Share this post


Link to post
Share on other sites

From the SR article:

Our 4K and 8k 70/30 tests showed performance being a small fraction of current-generation hard drives

Ouch... agreed with fallbreak, looks like something perfect for Amazon Glacier and not much else.

Share this post


Link to post
Share on other sites

This technique looks very suitable for archiving large media files - which are what the bulk of todays massive storage is used for anyway. To use them efficienctly I think the shingle-bands should be exposed to the OS/driver to handle things effciently. Some possible optimizations which come to mind:

- relax defragmentation by the OS

- bundle writes more liberally before pushing them to the disk

- rather write to a new band than squeeze data into an existing hole in a almost filled band (and cause lots of re-writes)

- upon file modifications: rather than always overwrite starting from where the modification happened, and in the worst-case overwriting the entire band, start at the beginning of a band if the modification is to be done in the fisrt 50% of the band. This should half the average performance hit

- align logical block sizes with band sizes to restore write speeds to almost normal levels, trading in some capacity (OK for large files)

And thinking about this.. how's the state of linear overlap between bits in HDDs? I've heard in BluRays they're already overlapping 8 bits sequentially and triggering / coding the data on the signal flanks. This requires more states than 0 and 1 to be distinguishable. Is this already being done in HDDs? Can it be done, or would there be some fundamental limitation? This could increase the linear density further, which yields nice STR increases. The drawback would probalby be more fragile data and of course more complicated read-out. But if it can be handled..

MrS

Share this post


Link to post
Share on other sites

I worry about what this will do to power safety. Imagine you need to rewrite a few k in the early part of a band. The drive has to rewrite the whole rest of the stupid thing. Now the power fails partway through the rewrite. Whoopsy. Well, I'm sure those other bits that have been assumed safely written and not touched in months weren't that important... :blink:

Share this post


Link to post
Share on other sites

I wouldn't think of running any critical system without a UPS anyway.

Share this post


Link to post
Share on other sites

This depends on whether you carry it around or not. If this is used to increase the capacity of 3.5" low-rpm mass-storage drives (for NAS, Backup etc.) the disk placement could be very static.

MrS

Share this post


Link to post
Share on other sites

Shingled magnetic recording is already used by tape drives and there it makes sense. I don't see a need for even slower hard drives.

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