Initial Experiences with Primo Cache on a Souped-Up Old C2D Laptop

[BonzaiDuck]

I started a thread in the OS forum to unravel an initial mystery about getting this old Gateway E-475M to recognize all 4GB of a new 2x2GB Crucial SODIMM kit I installed. The initial problem was quickly put to rest, but the discussion turned toward the compatibility of similar Crucial 4GB modules with the GM965 chipset.

A forum colleague verified that folks had indeed been able to configure 6 to 8GB of RAM on an assortment of these GM965-based laptops. I'm pretty sure I have a "latest" or "special" BIOS in the lappie: However it was "purposed" initially, it came with a T8300 processor -- not the T7xxx cores with which it was released by Gateway.

So I'm at a crossroads for gambling one or two Ben Franklins on the "big experiment."

The laptop is now fitted with an MX100 SDD (on an SATA-II controller). Without installing the trial Primo Cache software, I think the seq-read bench was around 260 MB/s. With Primo configured to use 1GB of the 4GB RAM, the same seq-read test showed about 350. The 4K tests were really quite astounding -- maybe as high as 120 to 140.

Somehow, I think the bottleneck -- excluding the SATA-II controller -- is in the processor and the Intel X3100 gfx accelerator -- which grabs some of the RAM. Since even with the active Primo Cache, the system shows about 1.5 GB of RAM available or free with a few applications running, I'm wondering whether I'll see any further improvements if I increased the RAM as I suggest above.

Ignoring the dollar outlay, I'm not sure if it's really worth it. On the other hand, despite the lackluster results with Primo Cache, I'm wondering if doubling or tripling the allocated cache would give me even higher scores.
I can order the RAM "just to see," and get a refund in the 30-day time-frame if it's only partially successful or otherwise proves to be of little advantage.

Any thoughts about this? Other folks with laptops might have a more robust experience to convey on this matter.

The nice thing: Primo Cache is pretty nifty, and especially nice because the "trial period" is 90-days -- not just a month . . .

ADDENDUM: I'm also wondering whether the lackluster results so far derive from the old DDR2-800 RAM limitations.

 

[Essence_of_War]

What exactly are you trying to do?

Messing around with RAM cache for an SSD will get you higher benchmark scores, but it's not going to make a real-world difference in how the machine runs.

So I'm at a crossroads for gambling one or two Ben Franklins on the "big experiment."

After an SSD, you'd have a very hard time convincing me to sink an additional 200 USD into a 7+ year old platform.

 

[BonzaiDuck]

What exactly are you trying to do?

Messing around with RAM cache for an SSD will get you higher benchmark scores, but it's not going to make a real-world difference in how the machine runs.


After an SSD, you'd have a very hard time convincing me to sink an additional 200 USD into a 7+ year old platform.

I think you're right about that -- definitely. Better to save the ducats for something better in the future.

I'd deferred -- even avoided -- fiddling with laptops for years. This refurb I picked up is a new toy. So far, I've sunk maybe $75 into the existing 2x2GB RAM upgrade, and $10 for the wireless-N mini-PCIe module. And add to that the MX100 -- which leads me to some other puzzling discoveries and ideas.

Like I said, I downloaded the trial Primo Cache to get some idea what would happen if I increased the RAM by another 2GB to 4GB. That is, I wanted to see how allocating a mere single GB to such a caching program would foretell what would happen with more RAM.

I configured it with 512MB and 1,024MB of RAM respectively. Now . . . according to the chipset/laptop spec, the storage controller is "AHCI" and "SATA-II" ready or "compatible."

But caching the MX100 in RAM barely results in as much as 100MB/s improvement in the AS SSD benchmark for sequential reads and writes. frankly, the writes improve to a greater degree.

It doesn't matter much whether the cache size is 512 or 1,024MB, either.

As it stands, without caching, The SSD seems to show about double the performance of a WD "Blue" SATA-III HDD. Since the throughput of the latter in the spec doesn't exceed 147 MB/s, you would think the controller wouldn't matter. It would only matter for the SSD.

But it is a mystery to me that the tests with Primo Cache don't show much of an improvement. Later today, I tried Primo Cache on an Intel P45 desktop system -- E8400 CPU -- and also 4GB or 2x2GB RAM, and also DDR2-800. There is a regular WD Blue 320GB boot-system disk in that box -- no SSD.

I used the CrystalDiskMark benchtest program instead of AS SSD: maybe that explains the discrepancy, maybe it doesn't. I used the same two Primo configurations. The benchmark shows in excess of 2,000 MB/s sequential read. How do I get less than 300 for the MX100 in the laptop? You would think the two different bench programs would still show fairly similar results, but they don't appear to do so.

The other thing I've dredged up from web-searches: You might not get any significant benefit of more RAM (exclude the possibility of a RAM-cache for the SSD or HDD.) If your applications use above 80% of RAM when multi-tasking with them, it might seem worthwhile to increase the RAM. But anything less than that, the prevailing wisdom says you might not get any improvement. I can load up COrel Photopaint, MS Word, Outlook and Paperport 12 -- RAM usage still doesn't exceed 50%.

But the puzzling thing is the miniscule improvement of SSD performance in the laptop with even as much as 1GB allocated RAM to Primo Cache. It almost looks like there is "just no improvement" or even a deficit, excepting the 4K or 4K-64thrd results.

If I could believe that the results on the desktop benchmarks for caching the HDD with 1GB of RAM would transfer to the laptop, it might then make sense to REPLACE THE MX100 with the WD Blue HDD that came with the laptop. It might make some slight sense to increase the amount of laptop RAM and see how far that goes for caching the HDD.

But maybe this is due to limitations of the laptop storage controller.

I think this is about as far as it goes for this laptop.

Certainly it's old technology, but it's quick enough for most things I'd want to do with it.

Time to stop fiddling around with it. No need to find ways to spend money, if no improvements can be made.

 

[BonzaiDuck]

Apparently the discrepancies I'd noted between the laptop with 4GB DDR2 and the desktop with 4GB DDR2, using Primocache -- whether with SSD (laptop) or HDD (desktop) is the benchmark test. You can't so easily compare AS SSD with CrystalDiskMark.

Whether it's just "benchmark" versus real world, or some improvement in actual performance (which I also observed), the results with Primo Cache were about the same using CrystalDiskMark across the board.

But whether or not it's worth the ducats to spend on 1x or 2x a 4GB SODIMM, that's another issue.

Primo Cache is pretty nifty. I can give up as much as 1GB of RAM leaving three for the OS and programs -- just sufficient with enough slack that you wouldn't be desperate for more RAM. Using 1GB versus 512MB for caching, though, is a fairly big jump in bench scores. About 1,600+ MB/s sequential read, and 1,400+ for sequential write.

Well -- so much for "old-tech" today . . .

 

[VirtualLarry]

I honestly don't see the point of RAM-based disk-caching software, above and beyond the OS-level file cache. Especially one that required a fixed, dedicated RAM allocation, taking RAM away from programs unnecessarily.

Edit: Save for running certain synthetic benchmarks, and being able to go "oooh, aaah" at the numbers produced.

 

[BonzaiDuck]

I honestly don't see the point of RAM-based disk-caching software, above and beyond the OS-level file cache. Especially one that required a fixed, dedicated RAM allocation, taking RAM away from programs unnecessarily.

Edit: Save for running certain synthetic benchmarks, and being able to go "oooh, aaah" at the numbers produced.

Not an uncommon point of view on this.

THE CONS:
It introduces an extra element of complexity to running processes/services etc.

It isn't "real" in the sense that there's no performance improvement for programs not previously (recently) used or in the cache.

THE PROS:

Definitely loads programs/services frequently used much faster -- differences are noticeable.

Takes advantage of hardware that otherwise may be "surplus" or idle.

THE PITFALLS:

Power outages without battery-backup, may lead to disk corruption, just as with BSODs and instability without the caching software.

Better be sure your RAM works flawlessly from the git-go.

. . . Especially one that required a fixed, dedicated RAM allocation, taking RAM away from programs unnecessarily . . .

Actually, correcting this misperception points up some positive features of the software. You can adjust the cache "on the fly," changing the amount of allocated RAM, pausing the cache, stopping the cache, etc. It seemed to me that no reboots were required, and it didn't matter what other programs were running. Another feature: the so-called "L2 cache" option. You can save a cache to disk so that it will reload at startup. This is definitely an improvement over Samsung's RAPID -- which has no persistent cache similar to the ISRT configuration between SSD and HDD, or the "Hyper Duo" feature of the Marvell disk controllers.

And -- unlike RAPID -- you can cache several storage devices at once. Some Primo Cache advocates (there's a forum) point out that you can use Primo side-by-side with RAPID, but note that simply using Primo and excluding RAPID makes more sense.

Then there's another thought about it. I've come to a conclusion that many of us err in our DIY building with too much RAM -- or more than we ever use. This was observed with RAPID [Magician 4.]: the RAM cache is so ample, that the benchmarks seem outrageously high. But somehow, there's ample RAM for all programs.

I think the rule of thumb about "enough RAM" had been suggested: "80% or more RAM usage means you could use more." So if you seem to show less than 50% usage all the time, you could choose to squeeze out some more "performance" by using the caching software. Also, for some XP, VISTA or Win-7 [EDIT] 32-bit diehards, it may be worth investigating whether the "unrecognized" RAM can be used by a caching program.

 

[BD2003]

Not an uncommon point of view on this.

THE CONS:
It introduces an extra element of complexity to running processes/services etc.

It isn't "real" in the sense that there's no performance improvement for programs not previously (recently) used or in the cache.

THE PROS:

Definitely loads programs/services frequently used much faster -- differences are noticeable.

Takes advantage of hardware that otherwise may be "surplus" or idle.

THE PITFALLS:

Power outages without battery-backup, may lead to disk corruption, just as with BSODs and instability without the caching software.

Better be sure your RAM works flawlessly from the git-go.



Actually, correcting this misperception points up some positive features of the software. You can adjust the cache "on the fly," changing the amount of allocated RAM, pausing the cache, stopping the cache, etc. It seemed to me that no reboots were required, and it didn't matter what other programs were running. Another feature: the so-called "L2 cache" option. You can save a cache to disk so that it will reload at startup. This is definitely an improvement over Samsung's RAPID -- which has no persistent cache similar to the ISRT configuration between SSD and HDD, or the "Hyper Duo" feature of the Marvell disk controllers.

And -- unlike RAPID -- you can cache several storage devices at once. Some Primo Cache advocates (there's a forum) point out that you can use Primo side-by-side with RAPID, but note that simply using Primo and excluding RAPID makes more sense.

Then there's another thought about it. I've come to a conclusion that many of us err in our DIY building with too much RAM -- or more than we ever use. This was observed with RAPID [Magician 4.]: the RAM cache is so ample, that the benchmarks seem outrageously high. But somehow, there's ample RAM for all programs.

I think the rule of thumb about "enough RAM" had been suggested: "80% or more RAM usage means you could use more." So if you seem to show less than 50% usage all the time, you could choose to squeeze out some more "performance" by using the caching software. Also, for some XP, VISTA or Win-7 [EDIT] 32-bit diehards, it may be worth investigating whether the "unrecognized" RAM can be used by a caching program.

You've actually got it backwards - it does have a persistent cache, it keeps a list of the files to be cached, and repopulates the RAM cache shortly after boot. Primocache on the other hand, does not. The other difference is that RAPID is based on caching frequently used files, whereas primocache is a dumb cache that only stores recently used file - again, nothing that the windows cache doesnt do already. Rapid also compresses the cache to make the most of the memory it uses, and caches writes in such a way that the EVO can use it most effectively. Even though few arguably need RAM caching for an SSD anyway, RAPID/Magician is a well designed piece of software, and primocache is a useless piece of garbage.

In other words, RAPID caches what you're likely to access, BEFORE you access it = good. Primocache only caches AFTER you access, redundantly with the windows cache = bad.

Primocache's ram caching is indeed almost utterly pointless. It just has the effect of bypassing otherwise uncacheable scenarios and making benchmarks look ridiculous, when the real world benefit is actually negative, since it's redundant with the windows cache. You're basically throwing memory away so you can get silly high numbers in a benchmark. The L2 cache on the other hand, would have a useful purpose for caching HDDs with an SSD....if only it wasn't so poorly designed. It just doesn't cache properly.

 

[BonzaiDuck]

OK . .. I get mixed results with it, anyway. Certain programs I use, which seem sluggish on this machine, actually benefit. The bench scores aren't all that good, and I'd attributed that to the storage controller.

There's no doubt even in my mind that RAPID is a much more polished piece of software. It's just that there's no application for it with any other drive than a single Samsung.

So I wonder if there aren't "alternatives" to Primo.

 

[BonzaiDuck]

I'm just doing some casual web-searches to answer my own question in the earlier post.

I came up with this, which is apparently a block-level caching program:

http://www.superspeed.com/desktop/supercache.php

The way Superspeed's Web-site is designed, it seems easier to get information about the server versions -- probably compatible with all after WS2003. Actually -- I saw a disclaimer for Windows 2000.

They seem to offer software to several levels of government-- or at least three groups of public institutions.

I'd be very interested if anyone has tried this program, and interested in summary appraisals.

 

[BonzaiDuck]

Somebody denied the assertion in the subject line or stated the contrary: that it was a file-level caching program. It then becomes a question as to which information is accurate. Here's the summary take on Primo-Cache from the AlternativeTo web-page, listing a plethora of RAMDisk and RAMCache programs:

"PrimoCache Overview

"PrimoCache is a supplementary software caching scheme that cooperates with system memory to provide data caching for disks. It improves system performance by transparently storing data into memory such that future requests for that data can be served faster. PrimoCache caches data on a logical block basis (offsets within a disk) while windows cache manager caches on a virtual block basis (offsets within a file).

"PrimoCache can also utilize the OS Invisible Memory which is wasted on 32-bit Windows when 4GB memory or more is installed."

Then there is the review at "The SSD Review:"

http://www.thessdreview.com/our-rev...t-13gbs-and-765000-iops-in-60-seconds-flat/3/

The "intel" about Superspeed LLC's Super Cache (or SuperCache, or Super-Cache:

I give a high probability that the software house fishes for government contracts at all levels: local, state, non-profit educational institutions (school districts) -- across the range around the Beltway.

Superspeed touts a Microsoft Certified Partner status.

Romex Primo-Cache has a 90-day trial period; Superspeed Super-Cache gives you 14 days. Romex wants $30 for a 1PC license and ~$79 for a 3-PC license; Superspeed wants $79 for a 1PC license, and you must give them the machine name for your computer which appears on the Windows networking dialogs for your LAN. Who does that? Why can't they simply take a hash-code from your hardware? OF course, they allow you to move the license from one machine to another. Do you have to "report" the new machine name?

Romex is based in Shanghai; SuperSpeed's headquarters is in Sudbury, Mass.

What does it take besides a greasy palm and a bunch of money to be a "Microsoft Certified Partner?" And did you ever buy licensed software out of Hong Kong or Taiwan with your credit card, and a month later find that WalMart called you about an attempted $600 purchase on the card which was made in London, England?

I gotta stick to next month's budget, and this is also part of my obsessive play-time computer hobby itch.

By the way. I'm running a Penryn dual core, GM965 chipset, and the trial version of Super Cache. The Intel SATA controller is AHCI compliant, supposedly spec'd to the SATA-II standard. I've replaced the WD Blue 500GB SATA-III lappie drive with a Crucial MX100. What one doth swap, one can swap again.

The Trial Version has a disabled lazy-write or write-caching feature, which becomes available when you shell out the seventy-nine clams:

CrystalDiskMark
Read Write
Seq 742.9 192.5
512K 711.5 189.9
4K 131.4 25.47
4KQD32 553.0 187.1

Super Cache is also a block-level caching program.

I don't know if I should worry about the cyber-army of the PRC. And in Richard Condon's "Manchurian Candidate," the character Yen-Lo exercises oversight of Ellie Islin and Raymond Shaw -- in the Northeast.

Who can ya trust these days?