Samsung RAPID mode

[BonzaiDuck]

Just to add an epilogue to this . . .

I'd begun to assume -- many months ago -- that Primo was "storage-mode agnostic." After I made the assertion again a few days ago, I wanted to double-check, because I really hadn't tested the proposition first-hand myself. There were forum posts at ROMEX in which posters had set up Primo for a RAID0 array, but were now wondering whether Primo implements TRIM. They seemed to be confused or oblivious to the fact that, if the BIOS version didn't support TRIM, neither would Windows -- nor would Primo. Romex is explicit: it allows TRIM to work, where it CAN work.

Meanwhile, I just decided to put a direct query to Eric Chen at Romex tech-support to confirm the "agnostic" feature, and here's the answer:

PrimoCache doesn't require a particular BIOS storage mode. It is based on Window volumes whose underlying disks can be in IDE, AHCI, RAID or other modes.

Please feel free to contact us whenever necessary.
Thank you.

Best Regards.
------------------
Eric Chen

So that settles it, and I'm willing to bet Z15CAM can at least see it working properly, even if he chooses -- in the end -- to abjure adding it permanently to his configuration.

 

[Charlie98]

Since you guys resurrected my OP, I have another question which you perhaps addressed... but maybe I'm not understanding... Some of this stuff you are talking about makes my brains leak out my ears.

IRST. What does it do and do I need it, particularly with an SSD... or, perhaps, especially with an SSD?

I no longer run RAPID, nor Magician except to check for FW updates or run a benchmark, but I have IRST installed and running on all 3 of my primary machines (all Z68 or B75 boards with a boot SSD.) I'm always looking to uninstall useless software but some of this stuff I don't know that much about.

 

[BonzaiDuck]

Since you guys resurrected my OP, I have another question which you perhaps addressed... but maybe I'm not understanding... Some of this stuff you are talking about makes my brains leak out my ears.

IRST. What does it do and do I need it, particularly with an SSD... or, perhaps, especially with an SSD?

I no longer run RAPID, nor Magician except to check for FW updates or run a benchmark, but I have IRST installed and running on all 3 of my primary machines (all Z68 or B75 boards with a boot SSD.) I'm always looking to uninstall useless software but some of this stuff I don't know that much about.

Well, Charlie -- I would "reassess" your need for the entire IRST package. Just to repeat -- a clarification: IRST is the Intel software and driver-set; ISRT is a feature it provides for "HDD acceleration through SSD-caching in BIOS RAID-mode."

On this sig-rig I'm using to type this message, I apparently uninstalled IRST when I kicked ISRT out of the house. Part of the reason I kicked it: I wanted to convert from RAID to AHCI mode to use RAPID.

So this system is entirely in AHCI-mode of the BIOS storage settings. And I think there were other reasons I wanted to use the Native Windows MSAHCI storage driver. If I recall, when I wanted a hot-swap disk capability for my bay-and-caddy setup, or just for the eSATA port on my front-panel, the Intel AHCI driver wouldn't raise a "Safely Remove . . " icon in the system tray.

That may have changed with a later version, but I don't suffer for using the native Windows driver. I have the same setup in my Server, using the MSAHCI driver configuring my drive-pool to a PCI_E Marvell-chip storage controller.

It was only after the RAID to AHCI transformation that I discovered Primo-Cache -- which I might have used with RAID storage mode AND the IRST software -- except for the "safely remove" problem.

If I had my mobo storage configuration in RAID mode, then I'd feel compelled to use the IRST software, because it provides a GUI for managing the arrays. What I'd do about getting back the "Safely Remove" icon for hot-plug-enabled drives configured to the controller (in RAID mode), would probably bring a freeware Japanese utility called "Hot Swap!" out of my "software resources" closet. That, too, works great, and it just "disappears" when you reconfigure for bringing back AHCI and the native driver. I'd just rather avoid using it, and I'm not interested now in RAID configurations, so . . . .

 

[BonzaiDuck]

Just to be fair, here's a new thread that takes up the topic discussed last year over "MSAHCI versus Intel drivers:"

http://forums.anandtech.com/showthread.php?t=2431391

Apparently, this means that you can squeeze more performance out of the Intel drivers, and they may not pose the problem I mentioned about "Safe Removal" of hot-swaps, offering even better GUI options for doing those things.

I'm tentatively guessing that it depends on (possibly) a need to update BIOS, install the drivers, etc. I can also see where a user (such as myself) might not want to bother.

In the end, a personal decision weighing inconvenience and other factors. Should you or shouldn't you? Do ya wanna or don't ya?

 

[BonzaiDuck]

Last of a triple-sequence post.

I should've realized this from the git-go, and it is really quite obvious from the Samsung Magician benchmark alone.

RAPID makes use of deferred writes if you indeed tweak the "Advanced Optimization" feature. This either makes use of the device's "write-caching" property or "policy," or changes it. So you'd expect -- and will confirm -- that your sequential write rate is only slightly slower than the sequential read-rate.

Magician seems to encourage you to do this, noting that it "gives the best performance" (not an exact quote, but that's the gist of it.)

I'm running some tests with Romex Primo-Cache, using the Anvil's Storage Utilities benchmarks to collect the results. Romex/PRimo actually defaults to "no deferred writes." So you find that the sequential write-rate for a Sammy 840 may bench around 455 to 460 MB/s with Primo RAM-caching and no deferred writes, while the bench result will actually exceed the 6,000+ MB/s read-rate with a 10-second write deferral.

I'm "composing" a new thread about this, with pictures (and circles and arrows on the backs of each one [Arlo Guthrie]). Just don't know how soon, when -- or even if -- I'll post it.

But for those who might get their panties in a bunch about "MSAHCI versus Intel Unified AHCI drivers," the 4K random read and write scores under all these caching schemes -- with deferred writes -- is "way up there:" about 700+ to 800+ MB/s just for RAM-caching an SSD-cached HDD, and no less than that for the "pure" Samsung 840 RAM-caching.

Whatever you argue for the advantages or benefits of caching schemes, they must have a seriously positive effect on mainstreamer usage profiles.

They sure as hell seem to by my perceptions -- whatever part of it is "placebo effect."

 

[Essence_of_War]

I should've realized this from the git-go, and it is really quite obvious from the Samsung Magician benchmark alone.

RAPID makes use of deferred writes if you indeed tweak the "Advanced Optimization" feature. This either makes use of the device's "write-caching" property or "policy," or changes it. So you'd expect -- and will confirm -- that your sequential write rate is only slightly slower than the sequential read-rate.

Magician seems to encourage you to do this, noting that it "gives the best performance" (not an exact quote, but that's the gist of it.)

I'm running some tests with Romex Primo-Cache, using the Anvil's Storage Utilities benchmarks to collect the results. Romex/PRimo actually defaults to "no deferred writes." So you find that the sequential write-rate for a Sammy 840 may bench around 455 to 460 MB/s with Primo RAM-caching and no deferred writes, while the bench result will actually exceed the 6,000+ MB/s read-rate with a 10-second write deferral.

This is what I've been saying since the beginning of the thread, was it not clear earlier?

http://forums.anandtech.com/showpost.php?p=36358002&postcount=23
http://forums.anandtech.com/showpost.php?p=36359505&postcount=25

 

[BonzaiDuck]

This is what I've been saying since the beginning of the thread, was it not clear earlier?

http://forums.anandtech.com/showpost.php?p=36358002&postcount=23
http://forums.anandtech.com/showpost.php?p=36359505&postcount=25

The limit of the SATA controller is only relevant as a barrier to be overcome by inserting a newer, faster hardware layer into the storage "pyramid."

The matter of Windows' own file-level caching is discussed sufficiently in the Primo-Cache explanations at the Romex web-site. [I see the web-page URL ends in the word "index," so look for the "Overview" tab for the product if the page as I see it isn't initially presented:

http://www.romexsoftware.com/en-us/primo-cache/index.html

If it's a matter of reading a "new" 4K-magnitude file, then -- sure -- you'll be limited by the controller and ultimately the device itself. So in using a caching solution or strategy, the benchmark will show a more optimistic view of things as opposed to "real-world." In reality, there will be a mix of performance instances in the user-profile.

Note that using Primo or RAPID or even ISRT for certain tasks with large files will not improve things on the first pass. Apparently, Romex and other robust caching solutions will sort this out, so you're not likely to clutter caches with DVR movie captures and so on. This latter point made me reticent about including my "media" disk in any caching scheme. I now find that WMC with "Live TV" and so forth actually seems to work with "greater robustness" from a slow SATA-II HDD (SATA-II controller port) cached to SATA-III SSD and then again to 1.5 GB of allocated RAM.

It would be absolutely imperative -- whether "read-only" or with "deferred writes" -- that all your hardware should be in tip-top shape. The write-caching, like ISRT's "Maximum" setting, presents a risk if any instability is added to your system.

Otherwise, I've turned off RAPID for my Samsung EVO boot disk (not the sig-rig, but a twin), and created two, 1.5GB RAM allocations: one for the Sammy boot-disk; the other for the SSD-cached HDD. If it's totally stable, one can argue "why not?" like the kid in the investment commercial. What I'm left with, with gaming programs in the mix, Media Center playing CNN, my web-page and e-mail open and a few other items I like cluttering my desktop -- is about 7.2 GB of "free" RAM out of the total 16. That's more than you have left with just about anything on an 8GB system.

So like the kid bounding up the steps of the brownstone -- "Why not?!"

 

[BonzaiDuck]

Here's another annoy sequentially-posted thought, though.

Hardly necessary to provide this background -- but I get those daily ads from the Egg, choosing to avoid dismissing them as SPAM or junk. With slightly less frequency -- from SuperBiz, Directron, Provantage, etc.

It's always a temptation, no less an inspiration.

So? Today, I see this "33%-off! Seagate ST4000DM000 4TB HDD Shell-Shocker!" ~$125 -- tooo-dayyy!! And sure -- I'd like to pick up one of those Bad Boys.

But I have a collection of WD 500GB SATA-II and SATA-III "Blacks" -- some new, a lot more reclaimed from older builds. I've got Blues in -II and -III flavors; I got Samsung F3 -II's. My server has an 8TB pool of 2TB Seagates. I really, really don't adamantly "need" a single 4TB drive, even for coveting. Not now.

These new HDDs -- I know -- are faster. But I can throw in any combination of the 500's and 1TB's of the SATA-II flavor, cable them to SATA-II ports, and cache them all to one SSD, then cache the whole enchilada to a GB or so of RAM. ISRT? You only pair a single SSD and HDD, with no RAM caching. Samsung RAPID? Only good for a single Sammy. Hyper-Duo? That may be different, but I don't want to add in the spare controller -- not now.

Primo? I can cache 16 HDDs to a single SSD, with the entire, single "task" cached to RAM. If I added the spare Marvel controller, I could cache across the mobo Intel controller and the Marvel.

Maybe the fascination will wear off. But I'm more fascinated this week than I had been through my 90-day trial and the last six months or so with my paid license. Nice!

 

[Essence_of_War]

It's a 5900 RPM drive. It's going to be pretty slow in random read/write workloads.

 

[BonzaiDuck]

It's a 5900 RPM drive. It's going to be pretty slow in random read/write workloads.

That's right! Never mind that I DON'T need it now. But if I did, it would hardly matter in the scenarios of use to which I would put it.

If I buy an SSD, I won't pick anything that has a seq-read-rate OR a write-rate much less than 500. If I buy an HDD -- I care not. I only care about reliability and capacity for those.

I can put roller-skates on that slug.