Slashdot Mirror
WD Explains Its Windows-Only Software-Based SSHD Tech
crookedvulture writes "Seagate and Toshiba both offer hybrid hard drives that manage their built-in flash caches entirely in firmware. WD has taken a different approach with its Black SSHD, which instead uses driver software to govern its NAND cache. The driver works with the operating system to determine what to store in the flash. Unfortunately, it's Windows-only. You can choose between two drivers, though. WD has developed one of its own, and Intel will offer a separate driver attached to its upcoming Haswell platform. While WD remains tight-lipped on the speed of the Black's mechanical portion, it's confirmed that the flash is provided by a customized SanDisk iSSD embedded on the drive. The iSSD and mechanical drive connect to each other and to the host system through a Serial ATA bridge chip, making the SSHD look more like a highly integrated dual-drive solution than a single, standalone device. With Intel supporting this approach, the next generation of hybrid drives appears destined to be software-based."
Yeah, that was a nightmare!
Linux already supports SSD-HDD caching with normal drives so if anything, it will probably already work or work with little changes. Otherwise, just pick up a tiny SSD and ignore this solution. http://bcache.evilpiepirate.org/
Nice troll, btw.
It isn't clear, exactly, from TFA what the drive will look like when you plug it in. Both components(the HDD and the SSD) apparently can function as SATA peripherals; but they are both behind some sort of bridge chip, type unspecified.
If the 'bridge chip' is just a reasonably generic SATA port multiplier, then an unsupported OS, or Windows without the driver, will just see two drives, the larger mechanical one and the smaller flash one. This would leave the way open for any OS with SATA and AHCI support to do whatever it prefers to get the best performance(on Linux, I assume that'd be at the filesystem level, with something like btrfs)
If the 'bridge chip' is some sort of proprietary oddity, and the vendor driver is required to even communicate with the flash portion(presumably at least some part of the drive will be visible as a normal SATA device, or booting without specific BIOS support would be a problem...), then that's pretty much worthless.
The iSSD and mechanical drive connect to
I believe I speak for the majority here when I say.... D'ARGH! KILL IT WITH FIRE NOW! This is yet another pathetic attempt by WD to marry it's crappy line of mechanical drives to SSDs in order to stretch their relevance out a little bit longer and keep them from having to retool their assembly lines and such to produce SSDs exclusively. Weeeell, good for you guys. But as my father would say: "Shit or get off the pot." Either switch to SSDs, and eat the cost, or stick with mechanical drives because they're cheap. But don't waffle and try to do both; You're getting the worst of both worlds then.
#fuckbeta #iamslashdot #dicemustdie
that uses features in the chipset is windows only, so should anyone be surprised that hardware or software these hybrid solutions are geared to the everyday PC user?
Sounds like Apple's Fusion drive.
Perhaps WD is releasing this with Windows support and hoping people can upgrade to their drives in OSX?
is this the 90s again?
... in all the big Fortune double-digits, that have their data centers overflowing with Linux servers.
The worst part of this, is that when WD goes bankrupt, as a result of this brilliant business strategy, there'll be even less competition in the HD market, which always means higher prices.
As an end-user, I'm NOT going to put up with a solution like this.
Even if it somehow performs better than current hybrid drives.
Even though most of my work is done on a Windows platform.
Hybrid drives are already a big compromise for minute gains.
Tying it to an OS choice?
NO FUCKING THANKS WESTERN DIGITAL!
In a budget situation I'd rather just put up with a competitor's hybrid or a plain old mechanical disk.
In a performance situation I'd rather just spring the extra cash for a real SSD. Better returns and more flexible.
Chas - The one, the only.
THANK GOD!!!
WD has taken a different approach with its Black SSHD
They'll have a lot more explaining to do, once some hacker, cracks the SSH password, starts pwning WD disk drives, and they begin to spew forth spam... :)
And its worth reminding people, that Windows already caches stuff in RAM, if you had 24GB of ram then it would be a lot faster cached, and the only gain with these drives is on startup and then not by much (since Windows arranges the disk so the common items are close together ready for boot).
So WD simply remind everyone why hard disk makers are struggling to remain relevant.
only in black?
There was an unknown error in the submission.
Seriously, if you don't want to use Windows, just don't buy this drive and buy yourself an SSD. No one is forcing you here.
That's correct. Non-fw cache will be more effective.
We don't know (yet), how WD drives work - will we see one whole block device or maybe it will have two SATA ports with two separate drives. I'm not sure they will support access to SSD to the some encapsulated stream.
There are Linux solutions like bcache and flashcache that can deal with cache. So, maybe it's time to use it and include into kernel?
"It feels like I'm at the Zoo when reading this thread - I'm frightened, but it's interesting" (c)
Yep, and flashcache (facebook).
But none of these are included in the kernel.
Picking up tiny SSD is not a solution for notebook or for the desktop - you need more ports, more money, more energy etc
"It feels like I'm at the Zoo when reading this thread - I'm frightened, but it's interesting" (c)
Uh, okay, whatever.
Guess I won't be buying one. Best of luck to those that do.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
they might call this the Western Digital 8.
WD just shot themselves in the foot .. many times over! People should boycott this company and their proprietary SSD technology.
At under $250 for a 256 Gig SSD, why would anyone buy a combination drive anymore? This would be like GM announcing that their next flagship green vehicle is powered by 150 horses.
The dm-cache device mapper target was added to the kernel in Linux 3.9. bcache is apparently on track for 3.10
Hybrid drives really are only good for laptops but SSD size is too small. I just use mount points which are even available under xp (i think) and later windows installs to map out the bigger files.
Wonderful idea. I can't wait to run right out and buy that. /sarcasm
For all intensive purposes, "whom" is no longer a word. That begs the question, "who cares"?
WD cheaping out and calling it a 'feature'.
I'll pass.. And stick with real ssd drives. they're cheap enough now for everything but bulk storage of shit i'll hardly ever touch.
Think secure boot. If the 'bridge chip' has a key, that only the trusted driver can supply, then with UEFI and "secure boot", they have just locked down the machine to windows only.
You are being MICROattacked, from various angles, in a SOFT manner.
The nice thing about Seagate HHDDs or SSHDs or whatever the companies what to call them now is that they just work. You drop it in a system, it works like a normal drive but faster. The flash works like cache on a RAID controller or the like. It just speeds things up.
With this, there's mucking about. Even if you could use it as separate drives, why would you want to? If I want to to just some small SSD storage and larger magnetic, I can. In fact I do. In my laptop I have a SSD for OS and apps and an HDD (actually one of Seagate's hybrids) for media and samples. My desktop is the same but more and larger drives.
It just seems silly to me. An all hardware approach seems much better and clearly doesn't cost that much as Seagate's drives are not expensive.
Do it in OS-agnostic firmware, you lazy shits.
If I run sshd on my sshd? What if I have a virtual machine that runs sshd on a virtual sshd on a computer with an sshd that also runs sshd?
Also other stuff (*BSD, solaris, etc) has had support SSD caching with ZFS for a few years now. Linux ZFS exists too but not built into distros yet, still it's very easy to install on many linux distros. I don't know what the current state of ZFS is on macs but they may be able to do this too.
1. No, 24GB RAM is $60, not $200. I just bought 4GB sticks of DDR3 for 300 baht retail, = 6*$10 = $60.
2. RAM is a lot faster than flash, and if you're buying these for speed then you want RAM because its faster.
3. The RAM cache is there anyway, if you have the RAM the HD with flash is cached in RAM regardless and the flash cache adds nothing.
4. The RAM can be used for other things by the OS.
So it comes down to, do you have a boot time problem and does that boot time take a LONG time, yet somehow fits into a small 8GB-24GB of flash?!
" but using RAM as HDD cache if you can use flash as HDD cache isn't going to save you any money"
It *both* saves you money *and* is faster *and* renders the flash in the HDD largely redundant if you have a decent amount anyway.
So I have 8GB RAM, I can see from Task manager I'm using 800Mb of it, the rest is all file cache. If I'm only using 800MB data+program, then my programs are LESS than 800MB and the files fit easily into the 7GB cache.
So the only thing it speeds up is boot time, and then only by the read difference between Flash and HDD, once Windows has optimized the disk so the boot files are close together.
Any time someone mentions the word "hybrid" I immediately think "hack." Hybrid solutions exist between an ideal approach and current technology, with the disadvantages of both and not enough benefits from either. They primarily exist so that early adopters can say they own something based on emerging technology "x" without spending money "y" required to do it right.
I'm sure this will offend someone, but the same applies to hybrid models offered by major car manufacturers. Ideal technologies are all-electric, hydrogen fuel cell, liquid gas, nuclear, etc. but they are not affordable, safe, reliable, or efficient; therefore hybrids exist for the masses. Given a few years, the cost of ideal engine technologies will drop and current hybrid owners will find themselves left with replacement battery cost (offset by fuel cost savings).
The same is true for drive technology - incorporate flash memory technologies *where appropriate* today, take advantage of modern mechanical hard drives with wicked low latency today, or simply wait until market forces drive down the cost. Hybrid drives are just not worth it.
Like make the OS learn what stuff we usually read from the drive, and keep that stuff handy. That would be so cool!
It's probability because there is no set life just a chance of failure. For example, with tossing one coin you have a 50% chance of getting one head and with two coins you have a greater chance of getting at least one head. With two components the low risk of failure per year gives a very long time for each separately for the time when half are expected to fail. Combine them and the chance for either of them to fail is combined so the time when 50% of the complete systems are expected to have some sort of failure is reduced to a lower time than either of the parts. :)
That's what you get for saying "that's not how it works" without stopping to think about it
Twice the productivity!
sysadmins and parents of newborns get the same amount of sleep.
Maybe that phrase needs to be updated a bit. In more ways than one.
In my computer, I have a Crucial M4 SSD for the boot drive and the more speed fasterness "crucial" apps. Then I have a WD Black terabyte drive for all the shit that doesn't need to be maximum possible speed.
Sometimes, I change my mind what needs to be faster than what is not at the forefront of my mission anymore. That's when I move my files around manually. Mind you, these are usually 8 - 20 GB of files or whatnot. This type of operation I do not want an un-brained background process to be performing at random times. If it picked the wrong time, I might drop FPS in an online match that was worth so many imaginary dollars to nobody at that precise moment...
Also, considering the choices MS has been making lately with windows, and the way they are treating OEM's it seems to be a very stupid choice for any OEM to hitch anything to windows. Putting you companies success in the hands of Balmer is a good way to destroy a company.
Remember those great Intel software-based modems back in 1993 that Intel had for Windows 3.1 that weren't supported by any operating system afterwards?
Remember how Intel sold that line of business and left us all hanging? Yeah.
Software based hardware like this is destined to be a one-trick pony. Use it in one system, and then it's stuck on that operating system for the duration. You'll be left in the lurch when the next version of you OS is released.
Go ahead and line-up to get screwed by WD and Intel. I'll skip this round of fleecing.
in your Ultrabook? Is there a research/poll out there that shows people buying Ultrabook and they want 500GB to 1TB of storage? Unless you have a lot of videos, I think 256GB is sufficient. Are there many people carry around their movies library?
With how much straight SSD prices have dropped over the past few years, I don't even really see much need for a hybrid drive. In 2011 I bought a 60gb ssd for $95 ($1.58/gb). Today, I can buy a better performing 500gb ssd for $350 ($0.7/gb).
Wow.. thanks FYI.
Since OSX and Linux already do this, maybe only Windows needed a driver.
me: ".if you have the RAM the HD with flash is cached in RAM regardless and the flash cache adds nothing"
you "not true"
If the file is in RAM, then the cache on the HD adds nothing. I already pointed out the boot scenario is the only difference there, you're defining reboots as somehow not boots!? Your suspend data is not frequent data and thus not in the cache anyway.
" For instance, if you do a long sequential read from the HDD, it won't cache the whole read"
It's a bit of a moot point, since the long read is going into *RAM* anyway, so there's enough ram for it, even if there isn't enough flash.
You're paying a lot at $26/4GB, but even so its still worthwhile, based on GGP comment "24GB of SLC Flash, from Intel, is ~$120", you pay an extra $30 for a faster cache that's more flexible. I assume you get it cheaper if buy more in bulk?
"You're ignoring writes."
Writes are done in the background anyway. They don't slow down the user.
"Assuming it's a write-through cache you could still see a significant write boost compared to just an SSD."
Except flash is very slow to write, so slow that Android won't let you write to flash on the foreground thread. On Windows writes are done async the call returns before the write has completed and it doesn't slow you down as a result. Given the slow flash is to write that's a good thing.
" the benchmarks I've seen of other SSD caching solutions have looked reasonably good"
Point me to them, I've never seen anything real world other than a boot demo, and then they compare it with an unoptimized boot. The benchmarks I've seen treat the read times in isolation, but of course if its in RAM already, the read never happens, so the real world read time is zero and there's no much optimization possible on zero.
Hey Western Digital! Go fuck yourselves! I have maintained a very long line of 'open' hardware. It runs very well on open software. "optimized for crapware" reads to me 'they paid us a bunch of money to run their crap'. And that's just fine. I will blacklist your 'black drive' and you will lose sales galore. None of your stuff will work with any of the 'linux-like' operating systems either (hello Android, Hello ChromeOS). If you market to a portion of the market, you should expect a portion of the market at best. Last time I checked, PC sales are on the decline. You would think they would market to everyone. They only want to market to a declining portion of the market, maybe your thing isn't any good. One thought I have in my mind is this: one group of people who buy drives by the truckload are those people serving stored data on the internet: search engines, social media, cloud services. I would be willing to bet that the number of drives they go through per month is not an insignificant fraction of the number of drives sold due to PC sales. The majority of those sites run Linux (ok, I'm thinking about Google, YouTube, Facebook, Twitter, Pinterest, Amazon, Ebay, Reddit, and a large host of others). A data center using drives with a MTBF of 100,000 hours and using 100,000 drives (1 TB each, nominal for total storage of 100 petabytes), goes through 1 drive per hour, 720 drives per month. If they serve 500 petabytes of data, then they go through 3600 drives per month. Marketing 'against Linux' means a loss of 3600 drives per data center per month per service. Assuming each service I listed (Google, YouTube, etc.) has just one data center, of 500 petabytes (which is a Lie!: Google has multiple data centers all larger than that), you loose sales of 8x3600=28800 drives per month. How smart is it to ignore the elephant in the room?
Commonly used files are preloaded by Windows both at boot and after boot from suspend. If your Opera files are frequent they are in the preload.
"(I use HL2 because it, and especially the episodes, benefit tremendously from being located on an SSD.)"
That's not what's on offer here, you don't get to decide HL2 should be in SSD, the caching algo decides which files in HL2 are cached based on how frequently you use it. Likewise Windows caching algo decides the same for RAM. In WD's case, it works with that Windows caching algo, so they're one and the same algo.
" If I cycle through a couple different data-heavy games then go back to the first one, "
Specifically you'd have to cycle through data heavy games that use MORE than 48GB in total (to ensure the 24GB of ram equivalent would be completely flushed), AND don't share common files, so for example the games don't reuse DX and drivers because they'd be in cache and would break up your long reads algo, AND the game data are arranged sequentially on the disk for the algo to detect it, AND you cycle through them.... playing each and every game in its entirety in a way that loads the files in the same order as they are on the disk (that long read requirement).
So HL2 requires 4.5GB of disk, to need all of it, you'd have to play the game completely through in a fixed order (based on the order of files on the disk if that's even possible), then you need to find 9 other games, that don't share files in common with HL2 and play those all the way through... and ... then if you did that, you'd gain the seek time (50ms?) of a hard disk on the next game load.
That's a tough sell.
"Lots of writes are synchronous, you know."
Flash is very slow to write, you need them to be asynchronous even for the SSD cache. Presumably WD stick some RAM on there too to cache the writes to flash.
shut the fuck up i'm not a troll, you're a fucking troll.
that software you linked to is not FREE as in speech.
Even Speech comes with a price.
Even if I ran windows I can't see having a driver for my hard drive. It should just work no matter what OS I am running. Sad and stupid. That's okay though as for me I don't think the hybrid is the way to go. SSD for the OS and external platter type for storage. Like most compromises this seems like the worst of both worlds.
As in Secure Shell daemon?
Get your abbreviations right, stupid WD. No wonder you have to develop Windows-Only.
"Actual SSD drives that you use as a substitute for HDDs write faster than HDDs."
The same flash chips are used in the internal memory of Android devices as SSDs, and they're not fast (and not connected across slow USB/slow SD Card). SSDs have a ram cache (65MB to 512 MB) to cache writes, it's that that's fast not the flash. This is why they stick a ram buffer on it because the RAM is faster.
http://www.newegg.com/Product/Product.aspx?Item=N82E16820147192
" offers 256 MB of Samsung DDR2 SDRAM cache memory", just to re-iterate, SSD makers add RAM cache because RAM cache is faster.
"...unless it's a synchronous write. Many programs explicitly issue sync() calls to flush the buffer cache to disk."
Flash is slow, see above.
"Congrats on your ass-talking, considering that you just admitted that you have no information to support your claims."
Your cache files are in RAM, the read doesn't happen.
Anandtech: "this benchmark writes 106.32GB. It's the load you'd put on a drive after nearly two weeks of constant usage. And it takes a *long* time to run."
i.e. they chose a scenario that can't ever happen, and if you had the RAM to make it happen in the real world, then that RAM would be used for file cache.
"...except as I've repeatedly told you and you seem fond of ignoring, there are many situations where the data won't be in RAM but will be in a SSD cache."
So its frequent enough to be in the SSD cache but not in the preload cache and this is not just after boot time? Wishful thinking. I covered boot time already, the period after boot and before the preload is the only time it would have data the equivalent ram cache wouldn't.
It's an unpleasant truth here, but the SSDs are a tough sell. The vendors promote boot times on laptops (laptop HDs are slow), and benchmark scenarios that don't happen in the real world.
A hard drive with a serial ATA host port, and the hard drive itself decides what to do as a cache. Host hard drive controls the slave hard drive which is inaccessible to the operating system. There you go. You can use any hard drive as a cache for another.
There's your one-hdd-is-a-cache-for-another scenario. Its two separable devices.
Sounds like WD knows they have to write the drivers for WIndows, but for Linux they will just let the community do it. Saves them money.
Heaps of people seem to think this is the same thing... but it is an extremely inefficient way of utilizing your space - if you have a 2GB file with only say, 100 MB of which is "hot" then you are moving the entire 2GB file.
A controller / driver / intelligent operating system can determine which BLOCKS in the file need to be cached, so in the above example, instead of consuming 2GB of SSD, the caching would consume 100 MB for that file - leaving the other 1.9 GB free to cache something else.
Also, you are not able to re-shuffle what is cached, on the fly while using the system as performance requirements change. A controller/driver or intelligent OS can do this. So, say you're in a game that wouldn't fit on your SSD due to you putting other stuff there, and it needs to frequently read/write from part of a data file as you traverse a level. The controller/OS can cache that hot part of the file while you're in the game, automatically.
If you were shuffling the data around yourself, manually you just couldn't get that optimisation.
I run: Windows, OS X, Linux, FreeBSD. Just because you have a hammer, doesn't mean everything is a nail.
I have a Thinkpad with the mSATA HD cache. It is a SSD of 24GB (I think) and is run by a windows software driver.
Trouble is, when the machine is quiet, suddenly 25% CPU is used by the cache as it jerks itself off or something. So the whiny little fan goes wild as the OS and the cache driver have a good time reorganizing the data. This drives me nuts. I have been thinking of pulling the mSATA drive just to stop it from messing around.
I think it is faster though. I like the concept of the now-discontinued 7200RPM ST750LX (SSHD). It has the cache on the drive and does not require a driver. I have it installed on another notebook and it seems to give the caching effect without the use of external drivers. I don't know why they are going away from that type.
Yeah, a drive that depends on any kind of OS besides the bare metal firmware on the board to which it is attached makes me uneasy. There are just so many more answers to the "what could possibly go wrong" question.
... whatever
If the thing is hidden then that's going to be problematic, but if both devices appear to the system as seperate disks then it will be similar to what we already have.
I'm seen a VERY similar system already in some Lenovo IdeaPads where there is a 1TB HDD and a 32GB SSD which both appear seperately and are quasi-RAID'ed together by some freaky Intel thing I've never seen before.
I discovered this while trying to perform data recovery on the drive and it complicated things massively, to the point where I had to step back and admit defeat for fear of corrupting the data beyond recovery.
If this is the case with the WD system, you would effectively be buying an SSD and HDD in a single unit which opens up some interesting possibilities;You could forgo the extra drivers and just treat them as separate drives, having fat bloated Windows on the SSD part to achieve sane boot times and then having large games and the swap file and stuff installed on the HDD portion.
At the moment the only way to do this is if you have a laptop with a miniPCIe or similar socket that will take a miniPCIe SSD and they are even more expensive than regular SSD's!
I personally prefer straight SSDs in laptops for exactly one reason: Impact resistance.
But for people who actually look after their laptops and need large capacity but don't want sacrifice boot times etc. then this would be a good compromise.
Plus you actually GET closer to the the advertised 500GB or whatever instead of some "Oooh, we use the metric kilo/mega/giga!" BS.
Kibi, mebi and gigalo can bite my shiny metal arse, Kilo and Mega etc data storage prefixes are ALWAYS power of 2! Alwaaayyysseee! (But only for data storage! Exception to the rule you know!)
Anyway, roll on memristor-based SSD's!
Sadly, we can rarely afford their silence... .
Um, no, not really. While Seagate doesn't explain their caching strategy, my guess is that unlike things like bcache, which cache at the block level, the seagate probably caches at the logical file level, which would explain why it's Windows only, they only implemented their algorithm for NTFS. My guess is they broke with tradition and implemented a disk that is aware of the logical layout of the files on it, instead of one that simply manages blocks(with file metadata just being another one of those blocks)
File level caching has a lot of advantages, for starters there tend to be a lot fewer blocks than their are files, meaning that the overhead imposed by the caching algorithm is a lot lower(flashcache for instance uses 500 megs of memory per TB of storage IIRC). Furthermore, by caching at the file level you can apply heuristic rules like not caching movie or music files, where performance is very unlikely to be critical etc.
Of course this all comes at a cost, having the disk actually know about the logical layout of the file system does break a lot of conventions and introduces a large # of potential issues....Trying to upgrade the file system on one of these things is probably not very pretty.
Monstar L
Nice post, but the article is about WD, not Seagate....
and don't forget the excellent EnhanceIO, https://github.com/stec-inc/EnhanceIO. This allows you to add cache to any device on the fly without and pre-formatting. For the storage tiering inclined people, similar performance gains can be made with btier, http://www.lessfs.com/. Mark from lessfs also has a great deduplication project as well called...well... lessfs.
Except these disks are more standard. They're basically an SSD and a HDD hooked to a SATA multiplexer (that lets you connect more than one SATA device to a SATA port. NOTE: Note all controllers support MUXes. Also, both drives share the bandwidth of the upstream port).
So plug this into a Windows PC and install the drivers, and two drives become one. Plug it into a Linux PC and you see two drives. Plug it into a Windows PC without drivers and again, you get two drives.
I would be concerned about how accessible my data was without the drivers. So you're using Windows and your data is partly on the platter and partly on the SSD; you reboot to an OS without the driver (i.e. the driver breaks when you upgrade Windows, you boot into Linux, whatever) - can you still get at your data. My guess would be that whilst the contents of the drives will be accessible as two independent drives, they will be in some undocumented format and therefore irrecoverable.
they'll have some tools or another for it. of course nobody is forcing you to buy these.
HOWEVER! current hybrid drives that figure on their own what to keep on the ssd are total bullshit that only work fooling repeated read speed tests. this actually makes a lot of sense, that the os can hint to keep certain files on the ssd.
world was created 5 seconds before this post as it is.
Similar question: Which would you rather have: software RAID or hardware RAID? On Linux, software RAID is usually faster, cheaper, more reliable, less buggy & fuller featured.
So yes, I'd prefer the drivers in my operating system rather than buried in some inaccessible firmware somewhere.
I suspect the 'disk' is still just a block manager (well, disk(s)) that doesn't have any awareness of the FS, just the driver half that makes it 'look' like one volume 'transparently'.
This is of course similar to 'fakeraid' with all the potential trappings thereof. If it presents as two disks otherwise or something similar in a standard way (and the driver is just to deliver product ahead of the standardized implemention being available) that might not be too bad.
XML is like violence. If it doesn't solve the problem, use more.
The OS can only beat a hardware solution if the OS is programmed CORRECTLY to do things in a smart way.
Which probably means the user setting the parameters correctly too.
How likely is that?
If the SSD part could be assigned to partitions on the disk and excluded from others, then 90% of the "smartness" required can be solved with that alone and in a method that is portable.
OS install partition gets SSD caching. OS doesn't put temporary or log files in the OS install partition.
File-level caching also has a lot of disadvantages. First and foremost, as you mentioned, it has to be tailored to specific file systems, like NTFS. For Linux and Unix OSs, there are so many file systems that porting file-level caching software to every one of them would be a nightmare. Linux/Unix has a generic block layer, so block-level caching software written for Linux/Unix is file system independent.
Second, the list of heuristics can easily spiral out of control. You might not want to cache a video file of a movie you're watching, but you probably would want to cache a video file of a movie you are editing. You have to know where the file came from, where it's going, how fast you want it to get there (i.e., priority), what programs are using it, etc. There's just too much contextual metadata that needs to be tracked; the more complicated the contextual metadata is, the more complicated the heuristics need to be. Also, statically defined heuristics may be inappropriate for dynamic environments.
Third, why cache an entire file if you're only working on a very small percentage of it? I have to admit, some file-level caches will only cache the "popular" portions of files, but in that case, why not just use a block-level cache and enjoy its better performance?
Also, I'm confused about "there are a lot fewer blocks than files," did you mean that the other way around? If you have a single 100 MB file, and the system's block size is 4K, you have a lot more blocks than files. There may be a smaller memory overhead with a file-level cache, but there's a greater CPU overhead due to all the contextual metadata and heuristic processing.
bio->bi_end_io(bio, error);
http://www.phoronix.com/scan.php?page=news_item&px=MTM2ODM
After being in development for more than one year, BCache was finally merged on Wednesday into the mainline Linux kernel code-base. BCache serves as an SSD caching framework for Linux by offering write-through and write-back caching through a newly-exposed block device.
TFA states: "With Intel supporting this approach, the next generation of hybrid drives appears destined to be software-based."
Intel supported 'Windows only' Itanium processors as well and how did that work out for them?
WINDOWS GOOD LINUX BETTER.
For large sets, this will be our guide even unto death, for the LORD will work for each type of data it is applied to...
Picking up tiny SSD is not a solution for notebook or for the desktop - you need more ports, more money, more energy etc
A great many laptops have mini PCIe slots. I have a Dell Inspiron 1720 (core 2 duo from May 2008) that has THREE of these (two of which are open; one holds a wifi card which I recently upgraded). You could pop an mSATA in either of the other two. BIOS might not see it correctly (one of the slots in the 1720 is labeled to support mSATA, and docs say the others won't work, but I don't believe it... I had to put the new wifi in the mSATA slot because my card was half length and that's the only place I could get antanna's to stretch to - wifi catcher switch doesn't work now, but linux sees and uses the new card just fine).
For the desktop... I don't know what you're talking about. You can fit two 2.5" drives in a 3.5" drive slot (or in the external 3.5" spot), or more in a 5.25" slot, or use a PCIx card solution, etc.
More energy? Really? The SSHD's have a flash drive and spinning platters in them. Having those separate *might* use a little more power, but that just depends on what drives you buy. The difference will be negligible at best.
Several things people:
1) These are Hybrid drives, NOT SSD.
2) Worst idea ever, wtf, lol!
3) Who actually buys a Hybrid drive anyway?
They are usually advertised as the "best of both worlds", but I (and I think most, which is why they are not popular) think they are more like the worst of both worlds.
Don't compare them to SSD, they are not SSD. Don't compare them to HDD, they are not a HDD. For the love of god don't buy them. It makes much more sense to buy an SSD for your system drive, and a HDD for your storage drive, like everyone that knows anything will tell you. Mashing the two together, will just give you a shitty compromise, that you can likely find for better/cheaper in a higher end traditional fast RPM HDD.
Not embedding your cache controller in firmware which is independent of OS, and not only making it dependent, but dependent on Windows, well that is just stupid. Not only are you limiting your market, but you are going to have to try and keep up with every stupid change that they do. I have no idea why they would do this, what would be a single benefit. Perhaps they save a dollar in not having to put the firmware memory on board? Who knows, crazy.
Does anyone know anyone who has bought one? Can they formulate a cogent argument as to why they did?
To Summarize: Stupid, but Irrelevant anyway.
That's good unless your operating system is inaccessible.
That's a much bigger problem with Windows than with Linux. You can easily mount md-raid drives from linux booted from USB or CD. I don't believe Windows supports booting off of anything other than a hard drive.
I would not want a windows-only software hard drive.
… don't buy WD drives!
Lol, I find it hilarious when marketing supersedes common sense and logic.
Only for Linux is 3.10 > 3.9
Why are open source development projects afraid of whole number changes? Are they afraid to commit to a major version? What miracle needs to be performed to get to LInux 4.0 within 10 years?
I haven't thought of anything clever to put here, but then again most of you haven't either.
Similar question: Which would you rather have: software RAID or hardware RAID? On Linux, software RAID is usually faster, cheaper, more reliable, less buggy & fuller featured.
You're joking, right? Last time I checked, the MD system wasn't big on adding drives / swapping drives for larger drives / replacing bad drives online with no disruption, and offers no way of visually indicating bad disks for local hands to remove. I also couldn't find any agreement on how the system should discover MD volumes, and there seemed to be some bizarre recommendation of having an extra device for metadata or something. In the continued absence of ZFS, I'll stick with HP's HBA RAID for now.
Yeah, I meant the other way around, nice catch.
Monstar L
totally agreed, i rn a 60 GB SSD with two 2TB data disks (one is rsynced to the other so I have an immediate backup of everything all the time). I built the box in 2007 with a core2 and have upped the RAM to 8GB, which helped a little, but that SSD, oooh that made a huge difference
Subversion of spatial scale luxury decoration ideas.
I think you misunderstand that such a thing is an unlikely use case outside of embedded systems deliberately designed to save power by not logging stuff to disk at short intervals. There's some linux laptop stuff aimed at the same thing and from what I read it was not a trivial exercise to cut down on all that disk activity.
SSD failure rates indicate far less reliability over time than RAM. In fact some mechanical disks appear to have lower failure rates than SSDs at this time which is why I spoke up against the mistaken idea of the above poster of some vast difference. Tom's hardware or similar will help if you want some up to date numbers.
Lol, I find it hilarious when marketing supersedes common sense and logic.
Self irony is fun, especially when not intended, as in your example above.
Only for Linux is 3.10 > 3.9
No, it's actually that way for every piece of software with a major.minor.patch versioning scheme in existence, since ever. It's just that you, personally, are dumb and/or ignorant, or possibly just trolling.
3.10 > 3.9
3.9 > 3.1.0
That wasn't so hard, was it?
It's a version string, it's not a floating point number.
If could be written as "Version 3, Release 10" in which case you would have no problem recognizing that "Version 3, Release 9" was the prior one. But remember, devs are lazy, and keystrokes are expensive, so shorthand notation just gives you the important (numeric) pieces, with a visually minimalistic delimiter.
This is even more apparent for projects with Major/Minor/Patch level designations. They are, once again, just using periods as a simple delimiter in a shorthand notation. Surely you don't complain that 2.3.32 is less than 2.3.6 do you?
Unlikely STX device is file aware (it's still a block level device). More likely they internally track LBA hotness and migrate frequently used to flash. If true, could be prone to partial cache problems and unnecessary migrations (flash wear concern). Seems like a file aware approach would be more efficient