SCSI HD Performance Slower Than IDE???

[Radboy]

HERE is a shot of the EZSCI5 Explorer disk cache tab. Notice that it doesn't say that Enable "the ability" of the write cache, or Disable "the ability" of the write cache. It says, "Enable write cache" & "Disable write cahe" .. or maybe Adaptec doesn't know what they're talking about .. you're talking semantics. Let's get real.

I never said there was anything called a write back cache - go read my posts again. I said it's called a write cache. You're the one you said "write back". Look at the picture again.

Read what it says under 'Description': Enabling SCSI write cache will improve the overall performance of your system .. maybe Adaptec doesn't know what they're talking about again. Perhaps u should shoot their Engineers a note & straighten them out.

No one said Sandra was the best benchmarking utility, but we can certainly tell when the numbers go up & when they go down. Guess it's just a coincidence that *all* write_performance numbers for resinboy's drive (buffered, sequential & random) went up dramatically (some more than doubled) after he enabled the write cache (just like Adaptec said it would).

I noticed a 'significant' differance after enabling the write cache - wasn't even hard pressed.

I think u may be confusing the write cache of a SCSI drive with the Troubleshhoting setting "Disable write-behind caching" found in Windows' File system properties.

The write cache does *not* revert after reboot. Notice that EZSCSI5 even lists the *default* setting for write cache.

I have used the IBM utility - it sux.

Seek, not access. My bad. How careless of me. Thx for the correction. Ave latency on a 15Krpm spindle = 2.0ms. THat would make ave access = 5.9ms on the X15, and (4.9 + 3.0) 7.9 for the 36LZX. Its still faster.

 

[Pariah]

"Notice that it doesn't say that Enable "the ability" of the write cache, or Disable "the ability" of the write cache"

It says enable write cache, which implies an ability. Not enable the write cache, like you said, which implies it as physical cache. There is a difference, but you're right about it being mostly semantics.

"I never said there was anything called a write back cache - go read my posts again"

They're the same thing. What does this feature do? It delays writes back to the disk in the cache.

"Guess it's just a coincidence that *all* write_performance numbers for resinboy's drive (buffered, sequential & random) went up dramatically (some more than doubled)"

No, it isn't a coincidence, it's just further proof of just how bad this benchmark is. Understanding how the feature works should tell you enough, that performance will not increase that greatly, especially in sequential writes where there should be basically no improvement.

"(just like Adaptec said it would)"

Ok, I'd like to see where Adpatec has said write performance will double under certain conditions with this enabled.

Here's what MS has to say about the feature:

Select to allow the drive to do write-back caching. This may improve write performance,...

 

[Radboy]

With all u know about SCSI hard drive technology, I'm surprised u couldn't be of more help to resinboy.

u should be an English teacher, cuz ur a master at semantics.

First you said "there is no such thing as a write back cache". Now ur saying "they're the same thing". Which is it? You sound confused.

You said, "It delays writes back to the disk in the cache." There is no "disk in the cache". There's a cache in the disk. And that's not semantics.

I (easily) notice a perf improvement after enabling the write cache on my SCSI hard drives (I have 3 of them - check the screen shot linked above), and notice a perf hit when it's disabled. It's easy to observe - regardless of what benchmarks do (altho they do go way up w/ the write cache enabled - just like Adaptec said they would).

What type of SCSI drive do u have that doesn't repond to enabling & disabling the write cache. Do u have a SCSI drive?

What kind of SCSI system? .. so we can see how it's done right?

I think ur confusing two different features.

You're quoting Microsoft. Microsoft doesn't make hard drives or SCSI adapters. Because u quote MS, I think you're confusing the write cache of a SCSI hard drive, with the write-behind caching feature of Windows. These are two (very) different things. Microsoft has nothing to do with enabling or disabling the write cache on a SCSI hard drive.

For example, if u go HERE to the VideoGuys (I do video-editing, and u learn a lot about this kind of stuff when u edit video), you'll find it says this (copy-n-paste):

By default, Windows 98 is set to enable disk write file caching. This function allows Windows 98 to write disk files in the background at some later time, rather than at the moment you initiate a disk write procedure. When recording video and audio files, the data is presented to the disk in one continual streaming operation, leaving no time for background writing operations. When write file caching is active, Windows tries to put the data in RAM first, waiting for an opportune time to write it to disk in the background. This opportune time never comes, and Windows runs out of RAM space, interrupts the recording trying to quickly empty its cache to disk, and potentially loses incoming data, causing glitches in the recording. To Turn Off File Write-Caching:

Click "Start" on the Windows Taskbar. Click "Settings" and choose "Control Panel."
Double-click the "System" icon.
Click the "Performance" tab.
Click the "File System" button.
Click the "Troubleshooting" tab.
Check the "Disable Write-Behind Caching For All Drives" checkbox.
Click OK to save your changes.

This guy is not talking about enabling the write cache of a SCSI hard drive. What this guy is talking about seems to be what you are talking about, but it is not what resinboy and I am talking about. When u talk about enabling the write cache of a SCSI hard drive, u wouldn't mention Microsoft. This is why I thing you're confused.

What SCSI hard drive did u say u had, again?

 

[Pariah]

I don't know what you are trying to so with the first part of your post, but be annoying, so I'll ignore that.

I own a Seagate X15 running off a Tekram DC390U2W card, though I don't see any relevance as this is not an exclusive SCSI feature.

"Because u quote MS, I think you're confusing the write cache of a SCSI hard drive, with the write-behind caching feature of Windows. These are two (very) different things. Microsoft has nothing to do with enabling or disabling the write cache on a SCSI hard drive."

I am not confusing anything. In win2k (possibly 98 as well, I don't use it enough to know), if you go to device manager and click on the properties of any hard drive there is a disk property tab with a "write cache enabled" box. This is not the same thing that is described in the link you posted.

I missed the link to the quote where Adaptec states you will experience significant speed increases up to double in some situations.

 

[Radboy]

Is ur WC enabled?

Have u ever toggled it?

Have u benched ur drive both ways?

The theory is simple. Writing to cache is (always) faster than the time it takes to put the data on the disk. (Do u dispute this?) That's why HDD manufacturers put a cache in their drives, even tho it costs extra, and even tho it could lead to file system damage or data loss. Seems like a lot to risk for something that you'd be "hard-pressed to tell the difference". And that's why ppl spend lotsa money to put more RAM in their systems. RAM runs at nano-second speeds, while HDDs runs at ms speeds - that's 1,000,000X faster (or slower, depending how u look at it).

If u let ur system move on to the next operation, once the data is in the cache, it will (obviously) be faster than if it has to wait for the data to get to the disk. I don't see why u have a problem with this.

Adaptec says this: "Enabling SCSI write cache will improve the overall performance of your system" (will, not maybe). Different systems will respond differently, but perf of *all* will improve (even yours). The numbers may vary, but you will *never* see a bench that indicates lower perf from enabling the WC on a HDD. This is also why some HDDs (like the X15) have an option to buy a drive w/ *more* cache. You can get the X15 w/ 16MB cache, right? (but it will cost u more money). Why does Seagate offer this option if it offers a neglible perf improvement?

RAM/cache is not a *little* faster than the disk. It's a *lot* faster.

Maybe u could post some benches for us w/ ur WC enabled & disabled.

 

[Pariah]

"Writing to cache is (always) faster than the time it takes to put the data on the disk. (Do u dispute this?)"

No arguement. The problem is that it needs to be written back to the disk no matter where it goes before that. You're not speeding up much of anything by writing to cache first, you're only delaying the inevitable. The theory of writing to cache first has nothing to do with freeing up the system to do other things. The idea is to minimize the number of required disk accesses per write. By queing smaller files in the cache you can write them all to the disk at once in the smartest order rather than doing random writes for each one. This will improve performance when it is properly utilized. The problem is that the drive will almost always have a full cache (which is also being used for read ahead cache as well) which means it has to all be written as soon as it arrives. Disk cache size will play a role in the effectiveness of this scheme, but only with significantly large differences in size. With the standard 2 and 4MB caches we see today, there will be almost no performance increase, as cumulative disk writes of fewer than 2 or 4MB are rare.

"This is also why some HDDs (like the X15) have an option to buy a drive w/ more cache. You can get the X15 w/ 16MB cache, right? - but it will cost u more money."

The 16MB Cheetah X15 was cancelled, so that should tell you what Seagate thinks about larger disk caches.

"Maybe u could post some benches for us w/ ur WC enabled & disabled."

I refuse to use Sandra to benchmark. Running ATTO there was no clear difference between the 2 conditions. I could post the dozens of numbers it produces at different settings, but I don't see the point as they were all basically the same (within a few 100KB/s under all conditions). I was going to run HD Tach as well, but the write test requires a clear drive, and I'm not going to format my main drive to prove my point.

 

[Radboy]

u never said whether u run ur X15 w/ WC enabled or not.

The X15 w/ 16MB drive is still listed on Seagate's website. See here:

http://www.seagate.com/cda/products/discsales/marketing/detail/0,1121,244,00.html

The guys at the storagereview.com - who know a bit about HDDs - have this to say about HDD cache (copy-n-paste):

The use of cache greatly improves performance of any hard disk, by reducing the number of physical accesses to the disk on repeated accesses and allowing data to stream from the disk uninterrupted when the bus is busy.

The external transfer rate is the speed at which data can be exchanged between the system memory and the internal buffer or cache built into the drive. This is usually faster than the internal rate because it is a purely electronic operation, which is much faster than the mechanical operations involved in accessing the physical disk platters themselves. This is in fact a major reason why modern disks have an internal buffer.

end paste

Note the word *greatly* before the words *improves performance*. You say, "almost no performance increase".

The following is from Seagate's glossary of terms:

Write Caching

With host-controlled write immediate, a write-complete signal status is returned when data is transferred to the drive buffer instead of waiting until the data is written to the media. The seek, latency and write times are cut out of the total command completion as seen by the host.

Notice the words "instead of waiting". Waiting means slower = lower performance. And when your talking about waiting for "seek, latency & write" compared to cache speeds, you're talking about a perf improvement in the area of (about) 1,000,000 times faster. That's significant.

So we have a difference of opinion. That's cool. I have enabled the write cache on my IBM Ultrastars, and (I think) you have disabled the cache on your X15. Sandra shows a large improvement in all write performance specs after enabling the WC on an Ultrastar, but that is only cuz it is a POS utility.

 

[Pariah]

I run all my drives at the default which is enabled.

"The X15 w/ 16MB drive is still listed on Seagate's website"

The drive has officially been dropped. Read through the Storage Review archives, there are multiple posts from people who contacted Seagate and were told the product will not be released.

"guys at the storagereview.com - who know a bit about HDDs - have this to say about HDD cache (copy-n-paste): "

I don't know where you got that from, but I have been a regular SR for a while and am in the top25 in posters. If you have spent any time on the board, you will know that the majority of posters don't lose sleep over the amount of cache on their drives. I doubt that came from a regular poster.

"Note the word *greatly* before the words *improves performance*. You say, "almost no performance increase". "

I said there will be almost no difference between 2MB and 4MB of cache, not between no cache and any cache. Search through SR's review index and look for the comparison between identical Maxtor drives with different amounts of cache. They performed almost exactly the same.

"So we have a difference of opinion."

I don't disagree with your theory or that cache by itself is significantly faster than a mechanical hard drive. The problem is that you are only seeing half the picture, how the data gets from the system to the drive. You don't seem to understand that once the data reaches the cache it isn't magically on the disk platters, it still has to be written and cache cannot speed a physical disk write. Theory vs reality is often extremely different. Yes cache can speed up performance, but simple fact of the matter is that the majority of data passing through the system receives no speed boost from hitting the cache, since it will still have to wait for the disk to write data to the platters.

 

[Radboy]

I got my info from here and here. They are not from posters. They are quotes taken from articles written by SR *staff* (who should know at *least* as much as posters) - whom u sem to disagree with. See for yourself (you'll have to scroll down past the box at the top).

We're not talking about perf diffs between 2MB cache & 4MB cache (unless u changed the topic somewhere along the way). We're talking about the perf benefits of enabling the write cache (which just so happens to be part of a 4MB buffer in the case of mr. resinboy) vs disabling the (entire) abililty to use the write cache (at all). It's really a simple concept.

I'm saying that the perf benefits of enabling the write cache on a LVD-rated IBM Ultrastar with 4MB cache are substantial. These are the only drives I have experience with. I have provided numerous links to sppt my point, and we have benchmarks from resinboy that concur. Can u provide even one link to sppt yours?

What I am seeing (and saying) is that when I enabled the write cache on my Ultrastars - which run my OS, apps & swap file, my system runs noticeably snappier. What are u saying?

 

[Pariah]

"They are quotes taken from articles written by SR *staff* (who should know at *least* as much as posters) - whom u sem to disagree with. See for yourself (you'll have to scroll down past the box at the top)."

Can you do a little research before posting? The guide you quoted was not written by the SR staff. On the top of every page it says "Reprinted, with permission, from The PC Guide."

"I'm saying that the perf benefits of enabling the write cache on a LVD-rated IBM Ultrastar with 4MB cache are substantial."

All I'm saying is you are wrong. You keep quoting this amount or 4MB making a big deal, when very little is cached on the drive. The majority or disk cache is in main memory, not the disk cache. I see no point in continuing this debate and correcting multiple inaccuracies in everyone of your posts, so we'll just agree to disagree on this topic.

 

[Radboy]

SR staff posted that information on their website. Surely you're not saying that the staff at SR posted information they disagree with, are you?

I use the 4MB cache number cuz that's how much cache is on the drive I (actually) *use*. I have experience with this drive, & can better talk about its perf characteristics. We can talk theory, but actual experience is better than (mere) theory. I'm basing my points on both real-world experience *and* benchmarks.

I don't see why u have your WC enabled if u see no perf benefits, and u take on risking data integrity. Seems hypocritical. If I thot enabling my write cache provided no perf benefits, I would disable it.

It seems that u disagree w/ me, resinboy, Adaptec, Seagate & the staff at the StorageReview. I still never saw a single link to sppt your claims.

Yes, I agree to disagree, too. And I enjoyed it. Thx for being game.

I think the main point is that resinboy is happy that he's getting max perf from his 9LZX - that's where all this started - remember?

 

[resinboy]

So, how 'bout them Mets?

( we're gonna kick it up a notch when I receive my new 36LZX on tuesday- if anyone is interested, I will post similar benches.