News for hackers. Stuff that matters.
Re:The 80/20 rule for storage systems by jajaja123 on Sunday December 14, 2014 @11:51PM (#8750717)
Irrelevant - storage is going to be all-flash pretty soon anyway!
Re:The 80/20 rule for storage systems by jude- on Monday December 15, 2014 @12:25AM (#8750795)
I doubt it. If your access patterns follow a power law distribution, why should you pay more for less space, if the performance boost gained by doing so (i.e. putting the "long tail" onto fast storage) is negligible? Moreover, the need for storage space certainly isn't decreasing, and the storage/cost ratios for disk and tape certainly haven't stopped increasing. If your access patterns follow a power law distribution, a multi-tiered storage system lets you find the optimal cost/performance trade-off.
Re:The 80/20 rule for storage systems by zaroth on Monday December 15, 2014 @01:42AM (#8750921)
Mostly reliability. I would never carry spinning rust in my laptop again, those days are gone. But also form factor, simplicity, consistency of end-user experience, and eventually solid-state will be cheaper than spinning platters in any case. Every graph I've seen of storage price with series for platters and flash show those lines eventually intersecting.
Re:The 80/20 rule for storage systems by bsder on Monday December 15, 2014 @02:31AM (#8751006)
> Mostly reliability. I would never carry spinning rust in my laptop again, those days are gone.

Exactly. Spinning rust drives are now far inferior to SSD's for reliability. No 2.5" spinning rust drive I have bought in the last 7 years has lasted for more than 2.

I have yet to have an SSD fail on me.

Re:The 80/20 rule for storage systems by mitchty on Monday December 15, 2014 @04:35AM (#8751246)
I've had 2 ssd's fail on me. To the point that the data is unrecoverable, ssd's fail worse than hdd's from my limited experience so far. Hope you have backups either way.

I also have at least 2 2.5" hard drives that are running on 3 years. Anecdotes make for poor data sets. (note one of those 2.5" drives is getting smart errors now after 12800 hours of use, so probably going to die in short order)

Besides, with spinning rust getting to 8TB now, hard drives are just taking the old place of tape in a proper backup solution. That is memory->flash->hdd->tape->optical (last two could possibly be the same layer).

Re:The 80/20 rule for storage systems by zaroth on Tuesday December 16, 2014 @01:22AM (#8756084)
Would you trust a drive to carry that kind of lifespan? I think solid state will ultimately make the rest irrelevant, and in just a few more generations time.

Then again, when storage and compute are integrated, even solid state should some day become irrelevant. So I just think 'hard drives' are going away, and 'store' will be a resource, like 'compute', and most likely implemented in flash for bulk store, and backed up, on separate flash.

Re:The 80/20 rule for storage systems by mitchty on Wednesday December 17, 2014 @02:24AM (#8761426)
Carry what kind of lifespan? Generally I don't trust anything anymore and have a 2 local copy + one offsite policy for data I care about.

If I don't have that I don't care enough about the data generally.

As for storage and compute you mean memristors or fpga like?

Solid state has a rather huge size difference to overcome. Hard to beat spinning rust at the moment and at least near future for bulk storage.

Re:The 80/20 rule for storage systems by jude- on Monday December 15, 2014 @01:21PM (#8753412)
Sorry, I should have been clearer. I was referring to the cost/performance trade-offs when building a storage tier. The choice of persistent storage for a laptop obviously presents different trade-offs.
Re:The 80/20 rule for storage systems by avem2 on Monday December 15, 2014 @01:50AM (#8750934)
Thats the exact point. Why pay for all flash in a storage system when the costs don't justify the usage. The article is pointing out that our applications don't use all of the storage allocated to them. Thus we really don't need a one medium for all but rather a tailored approach based on your workloads.
Re:The 80/20 rule for storage systems by akanet on Monday December 15, 2014 @04:43AM (#8751260)
You're missing the point. The author talks about it a bit, but even within flash there are wildly different levels of performance vs cost.
Re:The 80/20 rule for storage systems by zaroth on Tuesday December 16, 2014 @01:31AM (#8756097)
That's a great point. I think even within flash, a single drive shouldn't mix technologies. Trying to solve it with a selection algorithm can't add performance without complexity; 'behaves oddly at times' is not how you want to describe your storage subsystem.