Discussion in 'Article Discussion' started by bit-tech, 12 Oct 2017.
Now all we need is JAMR so we can have a bad MAMR JAMR with a HAMR.
What is "magenetic recording"? (subhead)
"Now long it will take the technology to reach the market..." (last paragraph). Were you very pessimistic Gareth or just a typo?
I'm not sure the spinning disk roadmap really needs to go that far out, when today SSDs are brushing that capacity point.
I'm sure if I could be bothered to put a graph together comparing capacity points and price per GB of flash and mechanical disk, they would cross well before then.
Both fixed - ta!
I'd be happy with SSDs being on the slower side (by "slower" I mean, similar speeds to modern HDDs) if the capacity was to go up significantly for the same price. They would still be "faster" because access times would be lower.
Personally I don't see SSDs hitting the same price as HDDs until HDDs become a "commodity" item where supply is heavily limited and thus price parity is reached through a combination of SSD price decreasing and HDD price increasing...
Since we're speculating, there are a few things I'm considering beyond simply the R&D and manufacturing costs of HDD vs SSD
I'm not considering personal storage at all here - the demand for personal mass-storage is going down and will continue to do so with data moving to cloud based providers. The smallest HDDs you can buy these days already vastly outstrip your average consumer's demand for storage, and it won't be long at all until "enough" SSD is at price parity with HDD - I don't necessarily mean the same cost/GB, but the same cost for a given requirement. Price parity per GB comes soon after this at a consumer capacity point - where SSDs at this capacity point will continue to drop, there's only so far HDDs can go - a 20TB drive offering a better cost/GB than a 1TB SSD is irrelevant to a consumer that needs less than half of the latter.
When we look at storage from a shared service provider point of view, whether its the likes of Netflix or Spotify, Google Docs or Azure, SoftLayer, AWS etc - we have a different set of benefits to SSD over HDD:
- As the capacity of HDDs goes up and up, so does the need for more wasteful erasure coding or duplication, so the cost per usable GB is impacted.
- Data reduction technologies that simply don't work on HDD on the basis of performance suit SSD very well. So if after de-dupe, compaction, compression etc, if we look at the cost per effective GB, for many data types, SSD is already significantly cheaper than HDD as a storage medium. (though other's benefit little if at all)
- The cost of powering and storing data on HDD is already significantly higher than SSD per raw GB, let alone per usable effective GB.
- Copy-based backups have all but disappeared from existence
- Things like GDPR are forcing people to get better at ILM in general, meaning getting better at keeping only what data needs to be kept. Now much of the data boom is coming from keeping things for analytics and data mining, which is a workload typically unsuitable for spinning media.
Really the only thing I can fathom still using spinning disk by 2025 is mass object-based storage for data resistant to reduction technologies (i.e. compressed media) and needing global duplication (offsetting the redundancy requirement as a factor) and maaaaaybe the odd HPC workload. The market for HDDs in time would be just a handful of organisations (or rather, the storage platform vendors that supply these organisation), the HDD manufacturers that haven't covered their arses against this (I'm looking at you, Seagate) will be in trouble, HDDs will become niche rather than ubiquitous impacting $/GB whilst the $/GB of flash continues to march further downwards.
I agree with you Mister_Tad, but the future you're describing won't become reality by 2025, more like 2035-2040.
We'll have to agree to disagree on that one.
The only parties I've seen/heard talking about HDDs in any positive way that far out are the ones making them.
Here's your graph:
SSDs are getting cheaper per-GB, but HDDs are also getting cheaper per-GB. Unless you NEED ultra-low-latency random access, then for bulk storage it makes far more sense to use HDDs than SSDs, and will continue to do so for the forseeable future as storage demands increase.
lolwut? Deduplication, online compression, parity and data integrity, etc are all techniques that predate the availability of enterprise SSDs. For example, ZFS benefits from using an SSD as an L2ARC volume, but for the bulk storage HDDs are more than adequate, and you can operate a zpool without an L2ARC drive at all (and just slapping in an SSD and turning on L2ARC to fill it can reduce performance if you're not careful). For operations performed on stored date, the limiting factor is the host system (mostly RAM capacity, but also CPU) not the individual drive performance.
The graph compares raw cost per GB. For bulk, non-compressible data and geographical distribution, there aren't many ways to get around this. Largely static data, too.
The other use case I mentioned was certain HPC workloads, where throughput of HDDs en-masse are likely to still be king for some time to come.
For anything else, raw capacity isn't as relevant - effective capacity and TCO are more important.
Of course they do, but the low latency random access of flash makes them far more feasible to use at-scale - no more caveats around performance assuming the compute is there to back it up.
Separate names with a comma.