SSHDs and next gen file systems in NASes

SomeGuy133

2[H]4U
Joined
Apr 12, 2015
Messages
3,447
The new SSHDs have 32GB now of NAND now and i was curious if they made 6TB 3.5in 7200RPM NAS versions would that help NAS/next gen file sytems at all or would it be an added cost with little benefit?

Just curious thats all. Always want to learn more you know :D
 
It's up to 32GB now? Awesome. I've had reasonable success with the 1TB 2.5 SSHD and the 4TB 3.5 SSHD, but that's been as standalone devices. I don't know how how they'd really perform within a mdadm or ZFS array.

I'd love to see how well they perform with UnRaid, especially as the parity drive. sadly.... time/money isn't in the cards at the moment.

Thanks for the heads up about the 32GB! Even the current 8GB cache is pretty slick for standalone drives :D
 
I tried a few SSHDs for laptops at work and the results were way less than stellar.

Compared to SSDs, SSHDs are absolutely horrible.

Pretty much all of the time, it was only slightly faster than a regular HDD.

They are completely pointless as far as I can tell.
 
This is where I hope to eventually see xpoint. Although I know this is not in Intel's current plans.
 
I tried a few SSHDs for laptops at work and the results were way less than stellar.

Compared to SSDs, SSHDs are absolutely horrible.

Pretty much all of the time, it was only slightly faster than a regular HDD.

They are completely pointless as far as I can tell.
It has to do with work flow i think. I do far to many random things and it probably would never help me but if you have a consistent work flow with small select of programs i could see how it could really help.

This is where I hope to eventually see xpoint. Although I know this is not in Intel's current plans.
i am thinking this is where next gen file systems/hybrid ones need to be designed to use various cache levels well.

You have RAM, maybe other levels of cache like spare SSDs, special cache on HDDs liek the SSHDs, and you could have more levels depending on where tech goes. It would be nice to see some real studies working on a real solutions to make better systems.

128GB of RAM, 2TB of SSD, 1,536 GB of SLC on 24 HDDs (assuming they throw 64GBon 8TB drives), and 192 TBs of HDD

That could be a decked out 24 bay NAS and I see lots of room for optimizations....just saying we are far from using this tech even remotely efficiently if we really look at a good step up of tech.
 
It has to do with work flow i think. I do far to many random things and it probably would never help me but if you have a consistent work flow with small select of programs i could see how it could really help.


i am thinking this is where next gen file systems/hybrid ones need to be designed to use various cache levels well.

You have RAM, maybe other levels of cache like spare SSDs, special cache on HDDs liek the SSHDs, and you could have more levels depending on where tech goes. It would be nice to see some real studies working on a real solutions to make better systems.

128GB of RAM, 2TB of SSD, 1,536 GB of SLC on 24 HDDs (assuming they throw 64GBon 8TB drives), and 192 TBs of HDD

That could be a decked out 24 bay NAS and I see lots of room for optimizations....just saying we are far from using this tech even remotely efficiently if we really look at a good step up of tech.

I think a much better AND cheaper solution is software such as MaxVeloSSD.

HDD is storage, SSD caches most used data, and you can also use RAM as cache at the same time. And unlike SSHDs, you can use whatever size of SSD you want to cache the HDD. Screw the lame low rpm SSHDs with 8-32GB of SSD cache. Gimme 128GB+ of SSD cache and 4GB of RAM cache for my HDD. If you buy the more expensive version, the max RAM useable goes up quite a bit.

Sooo much faster than crappy SSHDs, and way more options as well.

Worked great for me in Windows 7 and 8/8.1, but have been having trouble getting it running in Windows 10. The authors say that it works fine in Windows 10, so I think I may just need to do a wipe of my system and start from scratch.
 
I think a much better AND cheaper solution is software such as MaxVeloSSD.

HDD is storage, SSD caches most used data, and you can also use RAM as cache at the same time. And unlike SSHDs, you can use whatever size of SSD you want to cache the HDD. Screw the lame low rpm SSHDs with 8-32GB of SSD cache. Gimme 128GB+ of SSD cache and 4GB of RAM cache for my HDD. If you buy the more expensive version, the max RAM useable goes up quite a bit.

Sooo much faster than crappy SSHDs, and way more options as well.

Worked great for me in Windows 7 and 8/8.1, but have been having trouble getting it running in Windows 10. The authors say that it works fine in Windows 10, so I think I may just need to do a wipe of my system and start from scratch.

again my point was if they made a high end SSHD like a toshiba 6TB 7200 RPM drive or a fancy WD 8 TB 7200 RPM drive with that built in cache. additionally you could easily add SSDs on top too but again programs need to be made for this to work.
 
again my point was if they made a high end SSHD like a toshiba 6TB 7200 RPM drive or a fancy WD 8 TB 7200 RPM drive with that built in cache. additionally you could easily add SSDs on top too but again programs need to be made for this to work.

Lol, that's what I get for mainly responding to the first part of the post.

So yeah, the software does exists and works very well.

I still wouldn't bother with SSHDs. What happens when the SSD portion dies? Or how bad does the caching work if what you need to keep in cache is bigger than the SSD portion of the SSHD?
 
WDC did have a HD and SSD as a pair at one point. Was worthless pretty much from what I saw the reviews. Using SSD's to cache HD's are not worth it from what I saw. Even intel is not touting their caching chips any longer and neither is anyone else.. In time maybe with SMR and high density storage some kind of caching strategy would help.. But a RAM cache is always much faster.. You got to understand, the transfer rates are so fast now using DMA that any kind of break in the chain slows it down dramatically. Moving data in RAM itself is much faster.. So getting more RAM to use as a disk cache has more benefit than spending money on these caches. I think it was baidu that moved even SSD operations inside the computer rather than letting the SSD itself do it because it was faster and they had more control. They cache search data from their hard drives to the SSD which is controlled by the computer itself using RAM. A lot of work for an average user load but it helps when you are dealing with millions of requests. But the new file system will have routines that help with SMR drives and caching data.. Maybe an SSD there to cache such data might help as you can never add that much RAM to the system.
 
WDC did have a HD and SSD as a pair at one point. Was worthless pretty much from what I saw the reviews. Using SSD's to cache HD's are not worth it from what I saw. Even intel is not touting their caching chips any longer and neither is anyone else.. In time maybe with SMR and high density storage some kind of caching strategy would help.. But a RAM cache is always much faster.. You got to understand, the transfer rates are so fast now using DMA that any kind of break in the chain slows it down dramatically. Moving data in RAM itself is much faster.. So getting more RAM to use as a disk cache has more benefit than spending money on these caches. I think it was baidu that moved even SSD operations inside the computer rather than letting the SSD itself do it because it was faster and they had more control. They cache search data from their hard drives to the SSD which is controlled by the computer itself using RAM. A lot of work for an average user load but it helps when you are dealing with millions of requests. But the new file system will have routines that help with SMR drives and caching data.. Maybe an SSD there to cache such data might help as you can never add that much RAM to the system.
we are dealing with 128GB of RAM vs 2-4TB of SSD vs 100-200TB of actual data. There is plenty of room for a middle cache like an array of SSDs or built in cache for HDDs or a combination. In a fully built 24 bay NAS using 6TB drives that is a 1,152GB:1GB ratio of HDD to RAM. More RAM is not always possible and is also prohibitively expensive. It is 400 or 500 dollars for 64GB of ECC RAM IIRC vs ~400-500 for 2TB of SSD NAND.

Plus many systems 128GB of RAM is already the max RAM allowed.
 
You are not talking about the average user here.. or even a power user but for specific users with special needs. The average user dont even need a NAS setup let alone caching anything. Even for a power user, just what is he going to cache? I mean a cache only helps often used data and what kind of user would use 1TB of often used data that this setup would work? I mean I have 20TB of data but no caching would help speed that up. I tried a few realistic caching solutions but none of them helped much.. Windows built in prefetching works because those files are used very often. People dealing with large files would benefit from more ram than caching anyway as SSD speeds are still limited compared to RAM speeds. If you look at the user loads, there wont even be a number that can create a percentage for users requiring caching of 100TB of data.

Which is also why I said they use special specific software and hardware for some systems as they are far more efficient and faster than any generic solution. Even though the generic solutions are also very expensive. Large companies have found that even the expensive solutions are slow compared to what they can do internally.. Individuals trying to replicate this wont even come close in performance.. I think just moving the SSD algorithms inside the computer from the SSD speed up the system like 600%.. Thats a lot.. But another thing is, they have cascading SSD's which can be made even more efficient as the computer talks to all of them rather than individual SSD's acting independently.. Kind of like raid with SSD's but actual NAND management being done..
 
Last edited:
I've wondered too.

If the SSHD were smart enough, it seems that 32GB would help ZFS usage. I imagine a lot of metadata would fit, and the flash could buffer ZIL writes, so no SLOG device needed. But I'm just guessing. Wish someone would publish tests.
 
You are not talking about the average user here.. or even a power user but for specific users with special needs. The average user dont even need a NAS setup let alone caching anything. Even for a power user, just what is he going to cache? I mean a cache only helps often used data and what kind of user would use 1TB of often used data that this setup would work? I mean I have 20TB of data but no caching would help speed that up. I tried a few realistic caching solutions but none of them helped much.. Windows built in prefetching works because those files are used very often. People dealing with large files would benefit from more ram than caching anyway as SSD speeds are still limited compared to RAM speeds. If you look at the user loads, there wont even be a number that can create a percentage for users requiring caching of 100TB of data.

Which is also why I said they use special specific software and hardware for some systems as they are far more efficient and faster than any generic solution. Even though the generic solutions are also very expensive. Large companies have found that even the expensive solutions are slow compared to what they can do internally.. Individuals trying to replicate this wont even come close in performance.. I think just moving the SSD algorithms inside the computer from the SSD speed up the system like 600%.. Thats a lot.. But another thing is, they have cascading SSD's which can be made even more efficient as the computer talks to all of them rather than individual SSD's acting independently.. Kind of like raid with SSD's but actual NAND management being done..
multi user set ups would see a major benefit and nearly everyone has certain data that is more used than others.

If you ran steam off a NAS that could be one home user set up that would help. Active projects that are larger than RAM but again i understand your point but there is plenty of room for various use cases this would help. I could see this would help my future use case given the amount of this I plan on doing with my NAS/WS. Plex, Own Cloud, back ups, maybe some hosting, Encoding, and more. 64-128GB of RAM for >100TB is not a lot of RAM.
 
I would not use SSHD disks with a new filesystem like ZFS

The reason on reads
The "intelligence" of the SSHDs is quite low compared to a cache like the ARC or L2ARC cache in ZFS.
While the ARC is not only 100 x faster than the SSHD cache its far better as it cares about last accessed and most accesses.
So invest money into RAM

If you need more cache than you can afford or use on a mainboard, you can add an L2ARC of any size and performance (SSD, NVMe)

The reason on writes
All writes on ZFS are using the faster RAM as writecache for a few seconds automatically to collect small random writes and writes them then as a large sequential write. If you have a powerloss or crash the data in the cache is lost. This is the same with an SSHD if it uses write caching. But only with ZFS you can add a ZIL/Slog for secure sync writes. But again this requires that ZFS has full control what data is in cache and what is on disk.

So I would say, with ZFS its a very bad idea, useless and contraproductive.
Save the money, buy more RAM.
 
I would not use SSHD disks with a new filesystem like ZFS

The reason on reads
The "intelligence" of the SSHDs is quite low compared to a cache like the ARC or L2ARC cache in ZFS.
While the ARC is not only 100 x faster than the SSHD cache its far better as it cares about last accessed and most accesses.
So invest money into RAM

If you need more cache than you can afford or use on a mainboard, you can add an L2ARC of any size and performance (SSD, NVMe)

The reason on writes
All writes on ZFS are using the faster RAM as writecache for a few seconds automatically to collect small random writes and writes them then as a large sequential write. If you have a powerloss or crash the data in the cache is lost. This is the same with an SSHD if it uses write caching. But only with ZFS you can add a ZIL/Slog for secure sync writes. But again this requires that ZFS has full control what data is in cache and what is on disk.

So I would say, with ZFS its a very bad idea, useless and contraproductive.
Save the money, buy more RAM.

as i said earlier no drive exists with it that could be used and its a hypothetical if they made a good drive with it and if file systems were designed with it in mind.

It's up to 32GB now? Awesome. I've had reasonable success with the 1TB 2.5 SSHD and the 4TB 3.5 SSHD, but that's been as standalone devices. I don't know how how they'd really perform within a mdadm or ZFS array.

I'd love to see how well they perform with UnRaid, especially as the parity drive. sadly.... time/money isn't in the cards at the moment.

Thanks for the heads up about the 32GB! Even the current 8GB cache is pretty slick for standalone drives :D
its only the laptop ones now. I clicked on the link about 32GB versions.
 
My SSHD's have been great, but I read that over time they get slower and slower, so be wary. I may have not noticed because they are not my primary drive.
 
Basically most of these SSHDs are just hard drives with absolutely tiny SSD caches.
And in some cases, like WD, they're entirely software (not firmware) controlled. This makes them the Winmodem of hard drive/SSD tech.
Also, because these two separate storage technologies are managed by vendor-specific software or the drive controller, you really can't give a filesystem the kind of control that something like ZFS requires to prioritize access.

SSHDs are an okay compromise when you have a fixed budget and want better performance than just a hard drive and more storage space than a stand-aline SSD offers at a given price point. But they're a niche product, becoming more obsolete every year as SSD size goes up and $/MB comes down.
 
Basically most of these SSHDs are just hard drives with absolutely tiny SSD caches.
And in some cases, like WD, they're entirely software (not firmware) controlled. This makes them the Winmodem of hard drive/SSD tech.
Also, because these two separate storage technologies are managed by vendor-specific software or the drive controller, you really can't give a filesystem the kind of control that something like ZFS requires to prioritize access.

SSHDs are an okay compromise when you have a fixed budget and want better performance than just a hard drive and more storage space than a stand-aline SSD offers at a given price point. But they're a niche product, becoming more obsolete every year as SSD size goes up and $/MB comes down.
Yea but i am talking about the theoretical. If they were designed in top end hdd with zfs and snap raid would they help. I know currebt drives are trasg for that but if they were designed with that in mind it would be interesting
 
*DISCLAIMER: This applies to the old 8GB ones.....thanks to post I just learned about the 32GB versions*

You just have to understand a couple things about SSHD's to appreciate them.

1. They are completely Operating System agnostic, so they can only look at repetition to decide what to cache. It doesn't know if you are saving your English class homework or a config file for Windows. So it really can't help your daily actions. But if you have things that open frequently, then it can help with that.

2. They used to use 1 single NAND chip. It was 8GB, but it was 1 single chip. The entire method of SSD's speed is the result of parallel reads and writes. Your data basically gets fragmented across the drive on purpose (for several reasons) and as a result it can read and write to multiple NAND chips at once. That single NAND chip probably has a read/write of like 15-20MB/s. But the latency of grabbing small files is where it excels.

I have one in my Mac media server. All it does it run iTunes and some cron jobs for updating the catalog, thats it. For that, it is quite responsive. But start up something else and it does take a while (like a HDD). The best part is that I never have to wait for the hard drive to spin up, because everything it does is cached in the NAND. The hard drive still spins up, but I do not have to wait for it. Course I still have to wait for the ZFS pool to come up....1 disk at a time......
 
Let me add this

The Laptop SSHD with 32GB NAND flash is 25% faster compared to an 8GB version

To me it doesn't sound like the "upgraded" the drives. It sounds like it probably is the same price or cheaper to purchase modern 32GB single NAND chips then older 8GB single NAND chips. 4x the capacity and only 1.25x faster sounds like they are still using a single NAND device. So my previous observation would still apply. Even if it completely wrote to the NAND, you'd still be looking at 40-50MB/s bandwidth.


This is kind of big news too.

The other celebration is our new 7mm-thin 2TB Seagate Mobile HDD
 
Last edited:
Back
Top