Cache SSD+HDD (like Intel SRT) for old system?

[R0H1T]

Primocache works at the block-level, as the developers state in their website, as most storage caches do. There are also some cache products that work at the file-system level, but they are fewer in number.

Now, about the Device\Harddisk0\DR0 block device, it seems that it is the virtual disk that the Intel Software RAID driver created over multiple physical disks, and this would explain why you don't see any speedup.

Not sure about that since I've used this software extensively, when it was in beta for nearly 2yrs I believe.

I don't have it installed on my desktop with RAID0 SSD's but I did use on my laptop previously, see my earlier results for instance ~ http://forums.anandtech.com/showpost.php?p=36498929&postcount=8

 

[ashetos]

Not sure about that since I've used this software extensively, when it was in beta for nearly 2yrs I believe.

I don't have it installed on my desktop with RAID0 SSD's but I did use on my laptop previously, see my earlier results for instance ~ http://forums.anandtech.com/showpost.php?p=36498929&postcount=8

From this post I can see the part number (st1000lm024) of the 1 TB Seagate/Samsung HDD. It is evident that AIDA64 was ran on this HDD and hence the low numbers.

 

[R0H1T]

From this post I can see the part number (st1000lm024) of the 1 TB Seagate/Samsung HDD. It is evident that AIDA64 was ran on this HDD and hence the low numbers.

Yes it was, on my laptop, but even with 8GB RAM (recently upgraded to 16GB) I din't see the kind of numbers I was expecting rather hoping for. I may revisit the application if/when I upgrade the storage on my laptop, to SSD, & am definitely looking at RAPID or some other OEM caching software like Plexturbo provided I don't find Primocache too helpful at that point in time either.

Btw thanks for your informative (last) post on the previous page

 

[BonzaiDuck]

Yes it was, on my laptop, but even with 8GB RAM (recently upgraded to 16GB) I din't see the kind of numbers I was expecting rather hoping for. I may revisit the application if/when I upgrade the storage on my laptop, to SSD, & am definitely looking at RAPID or some other OEM caching software like Plexturbo provided I don't find Primocache too helpful at that point in time either.

Btw thanks for your informative (last) post on the previous page

It's always nice to see folks cooperating in these exchanges.

I'd made earlier reference to PrimoCache -- perhaps initial inquiries about it -- some several months ago when I'd first installed it as a 90-day trial. There was some contention that it was "not block-level caching," but block-level-caching is exactly what Romex and Superspeed ("SuperCache") promote and guarantee.

It was especially useful on my older laptop, which I mentioned. As a sort of "Swiss-Army Knife" of caching features, I would use it thoughtfully.

For instance, as much as I don't quite see the point of caching an SSD-RAID0, I also don't see the point of an SSD-RAID0. Surely, the argument for the latter hinges on an observation that it is "real-world speed without any caching gimmick." But the argument against using RAPID might also apply to the hardware RAID0 -- no human is really going to notice a difference between actual 530MB/s and 900MB/s in storage sequential throughput. Maybe, with certain applications and scenarios, it might matter a little -- I can't say. I can argue that I see some slight improvement with RAPID as opposed to no RAPID, but the difference isn't a matter of "hourglass" experiences.

Yet I DO notice quite a difference in HDD usage and access for the L2 feature, just as I was happy with ISRT using the Intel solution.

In my case, I was building a second machine for a gen2/gen3 processor. It wasn't one of my "planned" projects. I was almost planning to use a traditional HDD for boot/system, didn't want to spend more money on SSDs of acceptably large capacity (512GB), and finally purchased a couple 256GB EVO's (supposedly) on sale. I deployed one in the "system-in-progress."

So with my ISRT experience behind me, no desire to force RAID-mode for "everything" on one controller, I could see that I could put some games on an HDD with Primo's "L2" SSD-caching feature.

The bottom line follows, though. If the software works without flaw, you're no less prudent for using it, even for just "thinking" it improves performance. If it doesn't work without flaw, then it shouldn't be in your toolbox.

And BSOD's? If you're an overclocker, you should attempt to get past those experiences earlier in your computer-building project. If you plan to use RAPID or PrimoCache -- assure yourself that the hardware, OS and installed software are totally rock-stable.

 

[Phynaz]

I am not against DRAM caching, I'm just raising awareness of the issues.

Also, read DRAM caching is dumb, NTFS already does this. If you are to use a RAM-disk for a cache you should cache both reads and writes in write-back mode, or deferred writes as they call it.

I can't think of a single scenario where write-through DRAM caching is useful, when NTFS already does it. If you go that way, go all the way through!

First, NTFS doesn't do crap. It's a file system. Operating systems do caching, file systems don't.

Does windows do block caching? Does Windws do write coalescing? What's the maximum write cache size in Windows?

Perhaps now that you are less misinformed you will refrain from telling people what they are doing wrong

Edit
Or maybe not going by this thread when everyone disagreed with you.
http://forums.anandtech.com/showthread.php?t=2338595&highlight=

Seems you have a problem with not knowing as much as you think you do.

 

[R0H1T]

Yet I DO notice quite a difference in HDD usage and access for the L2 feature, just as I was happy with ISRT using the Intel solution.

That;s the thing, with an HDD installed you're better off using at least some form of L2, even a high speed USB flash drive will do, but for that I'll have to abandon the existing ODD & that's something I need. I'll however go for an SSD upgrade sometime this year after which I don't see there being any need for Primocache, unless one's looking specifically for some form of RAM caching &/or wants to minimize his SSD read/writes perhaps for endurance.

 

[BonzaiDuck]

That;s the thing, with an HDD installed you're better off using at least some form of L2, even a high speed USB flash drive will do, but for that I'll have to abandon the existing ODD & that's something I need. I'll however go for an SSD upgrade sometime this year after which I don't see there being any need for Primocache, unless one's looking specifically for some form of RAM caching &/or wants to minimize his SSD read/writes perhaps for endurance.

It all depends on either "consumer-preference" or "business-preference," storage needs and storage options. Lemme recap: I purchased my first SSD -- 60GB Patriot Pyro -- in 2011 specifically for ISRT usage. Only this year -- beginning January 2014 -- did I acquire an additional Chronos 60GB, 3x 512GB (Pro and 2x MX100), and 2x EVO 256GB.

But I have a lot of HDDs -- 1TB, 500GB, etc. Mostly SATA-II and -III, but there are even some IDE drives in my collection -- topic of another recent thread I posted.

If a person wanted 12TB of JBOD, pooled or "other" storage, he/she would spend roughly $4,000 to $6,000 for SSDs. Using WD Reds or Seagate NAS drives, one would spend at most $600. Are we purchasing speed? Or capacity? Or both? A business entity, no matter how prosperous, would answer those questions for itself before winding up the purchasing agent and sending him forth.

So a $30 to $80 software item like Primo may change the answers to those questions somewhat.

Also notice that Romex has a server version of the Primo-Cache product. If I have a server as an enthusiast, it would only matter -- perhaps -- if a larger number of people were accessing it as a web-server. Or if there were 100 people in my household.

Not conclusive, but inclusion of such a version tells me Romex is trying to "be serious."

 

[ZeroRift]

Does windows do block caching? Does Windws do write coalescing? What's the maximum write cache size in Windows?

Here's some old, but still relevant information on file caching behavior in Windows: http://msdn.microsoft.com/en-us/library/bb742613.aspx

That being said, most of your questions are answered by the MSDN link that ashetos posted.

Earlier in the thread, Cerb also posted a good illustration of this:

As a long-time modded Bethesda game player, plenty of RAM makes the most difference. My SSD lets me back up the game folder as I try mods easier, but it has made almost no difference in speed of loading, which is mostly CPU limited, just as when I was running a HDD. Big games might not use much RAM for their process, but they will take advantage of Windows mapping cached files for them, instead of hitting the drives, and doing that is still a few orders of magnitude faster than going to an SSD.

Even if you do the SSD cache thing, if he's got 4GB RAM, there's room to improve. IMO, get 2x4GB. If the PC has 2x2GB, and it plays nice with the old stuff, you got 12GB, if not, 8GB. 4GB may be enough, but enough is still a ways off from ideal.

To answer your questions directly:

Does windows do block caching?
Yes, in 256KB blocks.

Does Windows do write coalescing?
I suppose that Lazy writes could be considered a form of write coalescence, however I'm not sure this is precisely what you are referring to.

What's the maximum write cache size in Windows?
The design limitation is 16TB for modern version of Windows. Obviously there are more practical limits that come in to play, for instance, x86 computers are rarely equipped with more than 4GBs of memory, which leaves a typical cache size of, at most, 1 GB.

Sources:
http://blogs.technet.com/b/markrussinovich/archive/2008/07/21/3092070.aspx
http://msdn.microsoft.com/en-us/library/aa366778.aspx

 

[ashetos]

First, NTFS doesn't do crap. It's a file system. Operating systems do caching, file systems don't.

Does windows do block caching? Does Windws do write coalescing? What's the maximum write cache size in Windows?

Perhaps now that you are less misinformed you will refrain from telling people what they are doing wrong

Edit
Or maybe not going by this thread when everyone disagreed with you.
http://forums.anandtech.com/showthread.php?t=2338595&highlight=

Seems you have a problem with not knowing as much as you think you do.

NTFS explicitly handles the system cache, and NTFS is part of the OS. You can observe that FAT handles the file cache completely differently than NTFS, and you are right, it is the filesystem cache of the OS. I talked about the NTFS cache, not being accurate with my terminology, but then again many things I said are not pedantically correct. The meaning of the NTFS cache does not change anything about my arguments, only that I should be talking about "the system cache NTFS explicitly handles". My terminology is not 100% most of the time, but I hope I get my meaning accross.

About your random questions, and your personal attack/free psychological analysis, of course blocks are cached and writes are coalesced, there is no software write-back cache that doesn't combine writes anyway. I don't know about any tunable maximum write cache on Windows and I did not claim I know such a thing. Maybe it's different in XP, Vista, 7, 8, I don't know.

Maybe you don't like me, but I only talk about things I've tested or implemented. And if you'd like to know, in my professional life I've written parts of a kerne-level file-system with a custom DRAM cache, I've written a block-level write-back persistent device cache, I've written a SCSI HBA driver + the controller firmware for virtualized storage and finally a distributed shared block device over TCP/IP. Most in Linux, some in Windows. I don't use my experience as an argument in my posts, I always try to use logic and some fundamentals to pose questions. Honestly, I expected the discussion in the technical forums to be of a higher level, I tend to get disappointed very frequently here.

Look at this thread, I simply noted that you should be aware of the potential data-loss and all hell broke loose.

 

[Ichinisan]

Here's some old, but still relevant information on file caching behavior in Windows: http://msdn.microsoft.com/en-us/library/bb742613.aspx

That being said, most of your questions are answered by the MSDN link that ashetos posted. Earlier in the thread, Cerb also posted a good illustration of this:

As a long-time modded Bethesda game player, plenty of RAM makes the most difference. My SSD lets me back up the game folder as I try mods easier, but it has made almost no difference in speed of loading, which is mostly CPU limited, just as when I was running a HDD. Big games might not use much RAM for their process, but they will take advantage of Windows mapping cached files for them, instead of hitting the drives, and doing that is still a few orders of magnitude faster than going to an SSD.

Even if you do the SSD cache thing, if he's got 4GB RAM, there's room to improve. IMO, get 2x4GB. If the PC has 2x2GB, and it plays nice with the old stuff, you got 12GB, if not, 8GB. 4GB may be enough, but enough is still a ways off from ideal.

To answer your questions directly:

Does windows do block caching?
Yes, in 256KB blocks.

Does Windows do write coalescing?
I suppose that Lazy writes could be considered a form of write coalescence, however I'm not sure this is precisely what you are referring to.

I had 16GB DDR3-1600 in my i7 860 rig I built in April, 2010, I had hoped it would make up for the awful performance of a WD 2TB "Green" drive. That system was ab-so-lute-ly excruciating, and it was entirely due to the slow HDD. If Windows caches on a block level, it certainly was not utilizing the memory capacity in the way that it should.

 

[redzo]

I don't think the P55 supports Intel's SRT caching. Please correct me if I'm wrong.

My nephew lives far away from me in another state. I only see him briefly every few years. A while back, we built a Core i5 system for him. The first i5 CPUs had *just* been released. I don't remember the chipset of the Gigabyte motherboard. Would it be P55?

He has suffered with a WD 2TB "Green Power" drive most of this time he's had this system. I know...it should never ever be a boot drive. Using that system must be excruciating.

His system is having some problems and I'll definitely need to work on it the next time I see him. I'd also like to give him one of my unused 4TB drives with some kind of SSD caching solution.

Going with SSD-only is not an option. He plays lots games with very large installations, like Skyrim. His Steam library is huge.

I don't think my nephew is technical enough to manage what goes on an SSD and what goes on the HDD with a typical SSD+HDD system. That's why I think it would work best to have a caching solution where only 1 logical drive is presented to Windows.

What options do I have? Is there a software solution or an affordable hardware solution?

I read that Intel's SRT only supports caching up to 64GB. If there's a comparable caching solution, would it support 240GB? I'm thinking of upgrading to a higher-capacity mSATA drive for myself, so I might give him my 240GB SSD.

Also: If there is some comparable caching solution, could it also have onboard RAM for an HDD write buffer?

Thanks for any advice you guys can provide!

My thinking is quite simple: let's not add that useless complexity into our solution.

Options:

1. I wouldn't recommend a software RAM disk caching solution to someone non-technical. Under this situation you are exposed to data corruption due to power loss, due to future possible bad RAM or, even worse, due to buggy RAM caching software solution. RAM is quite expensive. In order to really benefit, you'll need a lot of it. Too much complexity for someone non-technical. Hard to debug/fix a potential problem from a remote location.

2. 250GB/64GB SSD hardware caching through intel RST. So you've eliminated those potential 1 problems like power loss, buggy software caching screw ups, but damn, such a waste of ssd performance. You'll never get rid of "what's used the most" thingy, that's what will get accelerated, youl'll never achieve the true potential of that ssd, you've basically moved some of 1.'s software complexity at hardware level.

3. The one I would've probably chosen under this circumstances. Drop an at least 250GB SSD into the pc case! Educate your nephew on how to get used to a limited storage space, how to organize his data so that it fits on a limited partition! Show him how to install his most disk demanding game/software on the ssd and how to uninstall it! Those are good things to know/learn anyways. They will do him good on the long run.

4. Hybrid flash HDD. By far, the simplest and straight to the point solution based on my understating of your needs. You drop it inside the pc case. That's it. What your nephew access's the most, that will open the fastest. Off course that it has many disadvantages, but let's not forget that it is the cheapest(based on capacity) and most easy to maintain solution.

I would never take into account 1 and 2 for myself now days, even if I consider myself technical. SSD prices have gone down. Intel RST is a useless technology now and a thing of the past, that past when a few GB SSD's where at current 128GB price levels.

 

[ZeroRift]

I had 16GB DDR3-1600 in my i7 860 rig I built in April, 2010, I had hoped it would make up for the awful performance of a WD 2TB "Green" drive. That system was ab-so-lute-ly excruciating, and it was entirely due to the slow HDD. If Windows caches on a block level, it certainly was not utilizing the memory capacity in the way that it should.

Well, a cache is only as good as it's hit rate, which makes it notoriously unreliable for enhancing performance. There are some Windows features that work to mitigate that, but it's never a guarantee.

Correct me if I'm wrong, but I think primocache only targets volumes / files you specify, thus enhancing the perceived cache-hit rate. Contrast that to native caching, which attempts to cache a little bit of everything.

My personal (unverified) observations indicate that even a best-case cache hit in Windows will still not return 100% of the file/data you are trying to access from memory, even if it's all stored in one, contiguous file and you have plenty of available ram.

Let's also keep in mind here, that the native Windows cache is not persistant. Windows does nothing in the way of L2 caching without 3rd party support (ReadyBoost notwithstanding). This means that if your underlying storage hardware is slow, it may take a very long time to populate the cache. Back in the XP era, I remember how systems seemed to run faster if you left them to idle for a while after startup. I'm guessing in retrospect that that was in fact due to a slow IO subsystem gradually caching common requests to memory.

Do you still have that old system? I would love to hear your comparison between native caching and primo cache. It'd be great to have a nice "low-water" mark anecdote.

 

[R0H1T]

Correct me if I'm wrong, but I think primocache only targets volumes / files you specify, thus enhancing the perceived cache-hit rate. Contrast that to native caching, which attempts to cache a little bit of everything.

My personal (unverified) observations indicate that even a best-case cache hit in Windows will still not return 100% of the file/data you are trying to access from memory, even if it's all stored in one, contiguous file and you have plenty of available ram.

Yes volumes, specifically partitions, & you can select them for read/write or read+write even individually as to the type of caching scheme you wanna employ for each partition IIRC.

Well even if it's not 100% depending on the workload it can be as close as (just over) 99%

 

[BonzaiDuck]

. . . Correct me if I'm wrong, but I think primocache only targets volumes / files you specify, thus enhancing the perceived cache-hit rate. Contrast that to native caching, which attempts to cache a little bit of everything.

My personal (unverified) observations indicate that even a best-case cache hit in Windows will still not return 100% of the file/data you are trying to access from memory, even if it's all stored in one, contiguous file and you have plenty of available ram. . . .

Depends on how you define "target," or a distinction between "target" and "source." So far, my most recent and by no means "preliminary" test of Primo-Cache arose from the situation of adding a 500GB HDD to a system with boot/system 256GB EVO. I loaded 11GB (GIG-abytes) of a Steam racing simulator to the 500GB HDD and then created the PrimoCache L2 cache on a 60GB Chronos SSD.

The cache hit-rate for the HDD is perpetually 99% so far, and game access almost seems to come from SSD-speed storage.

I'm familiar with the gains from ISRT, and they are typically 80% of the caching-SSD's spec speed -- provided the same SSD is connected to an SATA-III port. For the HDD, it doesn't matter what sort of port -- SATA-II is fine -- as long as it's using the same Intel controller.

But it is also good to resurrect or "refresh" the OP's original post. We've had a marvelous exchange with his thread, and perhaps we all benefit. However, the usage-scenario and "user"-scenario of the OP could limit the choices.

As someone just said, the nephew could "learn some discipline" with a $120 256GB EVO SSD. We've discussed all the caching options, and there's no requirement for ISRT (IRST SW and Intel chipset-specific HW). We've beaten that horse enough, too.

Certainly there's the possibility that nephew could manage more than one of the possibilities or options suggested. Maybe so? Maybe not.

I think the key thing to remember: the nephew's computer (which is i7/i5-etc. gen 1 or later) needs an SATA-III port and controller to really take advantage of any options -- caching or no caching. If not, there are quality, inexpensive controllers available -- at least a few offering the Hyper-Duo option as well as usefulness with PrimoCache. And I noted that my laptop system has only SATA-II, but the performance boost from a 3GB RAM-cache assignment can't be overlooked.

 

[ZeroRift]

Yes volumes, specifically partitions, & you can select them for read/write or read+write even individually as to the type of caching scheme you wanna employ for each partition IIRC.

Well even if it's not 100% depending on the workload it can be as close as (just over) 99%

Heh, fun stuff. Always satisfying to see hardware put to good use. :thumbsup:

 

[Ichinisan]

My thinking is quite simple: let's not add that useless complexity into our solution.

Options:

1. I wouldn't recommend a software RAM disk caching solution to someone non-technical. Under this situation you are exposed to data corruption due to power loss, due to future possible bad RAM or, even worse, due to buggy RAM caching software solution. RAM is quite expensive. In order to really benefit, you'll need a lot of it. Too much complexity for someone non-technical. Hard to debug/fix a potential problem from a remote location.

2. 250GB/64GB SSD hardware caching through intel RST. So you've eliminated those potential 1 problems like power loss, buggy software caching screw ups, but damn, such a waste of ssd performance. You'll never get rid of "what's used the most" thingy, that's what will get accelerated, youl'll never achieve the true potential of that ssd, you've basically moved some of 1.'s software complexity at hardware level.

3. The one I would've probably chosen under this circumstances. Drop an at least 250GB SSD into the pc case! Educate your nephew on how to get used to a limited storage space, how to organize his data so that it fits on a limited partition! Show him how to install his most disk demanding game/software on the ssd and how to uninstall it! Those are good things to know/learn anyways. They will do him good on the long run.

4. Hybrid flash HDD. By far, the simplest and straight to the point solution based on my understating of your needs. You drop it inside the pc case. That's it. What your nephew access's the most, that will open the fastest. Off course that it has many disadvantages, but let's not forget that it is the cheapest(based on capacity) and most easy to maintain solution.

I would never take into account 1 and 2 for myself now days, even if I consider myself technical. SSD prices have gone down. Intel RST is a useless technology now and a thing of the past, that past when a few GB SSD's where at current 128GB price levels.

Option 3 means his high-end games installed on the HDD won't have any performance improvement at all. That's the primary use for his system.

With option 2 (or any SSD+HDD read cache option), the game data that's accessed most frequently will be cached. The multiplayer maps will load quickly. If he never plays the single-player campaign, those videos and maps will not occupy SSD storage space. Great! Nearly all the important parts of the OS would quickly end-up on the SSD cache. Great!

Option 4 is unappealing because I'd have to buy a pricey drive I don't already have; and they all seem to have a pitiful 8GB of flash. I'm already surrounded with extra 2TB/4TB HDDs right now. A 4TB drive with 240GB SSD cache should offer nearly all the benefits of an SSD and all the capacity he needs for games.

Option 2 is where it's at, but need an alternative to Intel SRT. PrimoCache might be it, but I'm thinking of getting a Marvel controller with HyperDuo. They have these with mSATA so my hand-me-down 240GB SSD would sit right on the controller card and a 4TB drive would connect to the same controller card via SATA3.

 

[redzo]

Those "magical" numbers can be achievable by using samsung's rapid which comes at no cost. Samsung's rapid is also "magic". It is pure trickery with very few cherry picked benefits, but you at least get assurance that worst case scenario is back to ssd. With software solutions under this scenario you're back to crappy mechanical.

Don't forget that caching principles stay the same no matter the implementation. No matter if it's at hardware or software level, you can't get rid of its drawbacks.

 

[Ichinisan]

Well, a cache is only as good as it's hit rate, which makes it notoriously unreliable for enhancing performance. There are some Windows features that work to mitigate that, but it's never a guarantee.

Correct me if I'm wrong, but I think primocache only targets volumes / files you specify, thus enhancing the perceived cache-hit rate. Contrast that to native caching, which attempts to cache a little bit of everything.

My personal (unverified) observations indicate that even a best-case cache hit in Windows will still not return 100% of the file/data you are trying to access from memory, even if it's all stored in one, contiguous file and you have plenty of available ram.

Let's also keep in mind here, that the native Windows cache is not persistant. Windows does nothing in the way of L2 caching without 3rd party support (ReadyBoost notwithstanding). This means that if your underlying storage hardware is slow, it may take a very long time to populate the cache. Back in the XP era, I remember how systems seemed to run faster if you left them to idle for a while after startup. I'm guessing in retrospect that that was in fact due to a slow IO subsystem gradually caching common requests to memory.

Do you still have that old system? I would love to hear your comparison between native caching and primo cache. It'd be great to have a nice "low-water" mark anecdote.

I sold that and built something new for myself. However, my nephew's system has the same horrible drive and the same motherboard chipset (P55). Knowing it will probably be the last opportunity we'll ever have to upgrade this particular system, we might upgrade him to 16GB RAM when he comes to visit.

 

[R0H1T]

Those "magical" numbers can be achievable by using samsung's rapid which comes at no cost. Samsung's rapid is also "magic". It is pure trickery with very few cherry picked benefits, but you at least get assurance that worst case scenario is back to ssd. With software solutions under this scenario you're back to crappy mechanical.

Don't forget that caching principles stay the same no matter the implementation. No matter if it's at hardware or software level, you can't get rid of its drawbacks.

The only drawbacks, the ones you're overly exaggerating, are that you're held back by the speed of your HDD, which in my case was a 5400rpm drive, especially without a fast L2 cache. The rest is, as you previously stated, purely magic :awe:

If setup properly Primocache is easily the best dedicated caching software I've seen ever, you'll still be limited by the slowest storage disk in your system (probably the one you're caching) but in the meantime your system will not have to constantly seek/access data from that drive greatly improving its responsiveness.

 

[Phynaz]

Here's some old, but still relevant information on file caching behavior in Windows: http://msdn.microsoft.com/en-us/library/bb742613.aspx

That being said, most of your questions are answered by the MSDN link that ashetos posted.

Earlier in the thread, Cerb also posted a good illustration of this:


To answer your questions directly:

Does windows do block caching?
Yes, in 256KB blocks.

Does Windows do write coalescing?
I suppose that Lazy writes could be considered a form of write coalescence, however I'm not sure this is precisely what you are referring to.

What's the maximum write cache size in Windows?
The design limitation is 16TB for modern version of Windows. Obviously there are more practical limits that come in to play, for instance, x86 computers are rarely equipped with more than 4GBs of memory, which leaves a typical cache size of, at most, 1 GB.

Sources:
http://blogs.technet.com/b/markrussinovich/archive/2008/07/21/3092070.aspx
http://msdn.microsoft.com/en-us/library/aa366778.aspx

My questions were rhetorical.

Anyway, Windows does not cache block operations. Windows file system cache is a file cache only.

Windows does not do write coalescing nor support TRIM In the cache.

Windows write cache on desktop systems is 1GB.

thanks for the links, interesting reading.

 

[redzo]

Option 3 means his high-end games installed on the HDD won't have any performance improvement at all. That's the primary use for his system.

With option 2 (or any SSD+HDD read cache option), the game data that's accessed most frequently will be cached. The multiplayer maps will load quickly. If he never plays the single-player campaign, those videos and maps will not occupy SSD storage space. Great! Nearly all the important parts of the OS would quickly end-up on the SSD cache. Great!

Option 4 is unappealing because I'd have to buy a pricey drive I don't already have; and they all seem to have a pitiful 8GB of flash. I'm already surrounded with extra 2TB/4TB HDDs right now. A 4TB drive with 240GB SSD cache should offer nearly all the benefits of an SSD and all the capacity he needs for games.

Option 2 is where it's at, but need an alternative to Intel SRT. PrimoCache might be it, but I'm thinking of getting a Marvel controller with HyperDuo. They have these with mSATA so my hand-me-down 240GB SSD would sit right on the controller card and a 4TB drive would connect to the same controller card via SATA3.

So your're thinking at replacing intel rst with a ram caching solution? If that's the case, I would chose a hardware implementation of caching(intel rst) instead of software anytime.

 

[BonzaiDuck]

My questions were rhetorical.

Anyway, Windows does not cache block operations. Windows file system cache is a file cache only.

Windows does not do write coalescing nor support TRIM In the cache.

Windows write cache on desktop systems is 1GB.

thanks for the links, interesting reading.

" . . . nor support TRIM in the cache . . . "

This had always left a thought in the back of my mind with ISRT especially with chipsets (my Z68) for which earlier BIOS did not support TRIM in RAID. But I never had any real trouble with it for three years.

Windows RECOGNIZES SSD-cache-drives, but only in the Device Manager under the "Disk Drives" node of the tree. Obviously, caching drives aren't meant to show up under Windows Explorer. I would think that TRIM is enabled for any SSD under hardware/BIOS AHCI mode, howsoever they are deployed.

 

[Ichinisan]

Those "magical" numbers can be achievable by using samsung's rapid which comes at no cost. Samsung's rapid is also "magic". It is pure trickery with very few cherry picked benefits, but you at least get assurance that worst case scenario is back to ssd. With software solutions under this scenario you're back to crappy mechanical.

Don't forget that caching principles stay the same no matter the implementation. No matter if it's at hardware or software level, you can't get rid of its drawbacks.

I really see no reason to bother with RAM caching in a pure SSD system. For benchmarks? On a pure SSD system, the SSD is plenty fast enough for my every-day use.

My nephew's scenario requires massive capacity, which SSDs just don't offer at a reasonable cost. I'm just hoping to give him 240GB of SSD caching to boost the performance of a slow, mechanical HDD.

 

[ashetos]

My questions were rhetorical.

Anyway, Windows does not cache block operations. Windows file system cache is a file cache only.

Windows does not do write coalescing nor support TRIM In the cache.

Windows write cache on desktop systems is 1GB.

thanks for the links, interesting reading.

Windows caches blocks just fine. If you're talking about raw block device access then it's a diffrent question.

Windows combines writes just fine, just do a 4KB sequential write workload on a file.

Windows supports TRIM in buffered IO since Windows 8:
http://msdn.microsoft.com/en-us/library/windows/desktop/hh848053%28v=vs.85%29.aspx

Edit:
Just did a test to confirm the maximum dirty buffer limit. Don't know where you got your 1GB number but it might be a much older version of Windows. On Windows 7 my Modified memory field in Resource Monitor grew over 2GB copying an .iso file.

 

[BonzaiDuck]

So your're thinking at replacing intel rst with a ram caching solution? If that's the case, I would chose a hardware implementation of caching(intel rst) instead of software anytime.

As the OP says, he's interested in an SSD-caching, HDD acceleration option.

Primo-Cache offers either RAM-caching or SSD-caching (the OP's interest), and it offers them separately or in combination.

I'm suspicious about claims that either Intel or Samsung offers a "hardware" implementation that is "better." What they offer in effect is a proprietary solution that cannot be used with any other controller or any other SSD. But they both require software in conjunct with their "unique hardware."

As the OP suggests with his most recent post, he won't choose the SSD part of the system based solely on the offer of a solution like Samsung RAPID, and he's much less interested in that sort of a feature than he is in using an SSD to cache a large HDD.

The options are many and clear. He only needs to decide and choose . . .