Christer

Comparing Hdds - Strange Results

Recommended Posts

Hello all!

I have two HDDs, a new 120 GB Hitachi 7K250 with 8MB cache and a three years old 40 GB IBM 60GXP with 2 MB cache.

The 7K250 is Primary Master and the 60GXP is Primary Slave.

(A CD-R/RW is Secondary Master and a DVD-ROM is Secondary Slave.)

The rearmost partition on both HDDs (16-18GB) are FAT32, dedicated to Ghost Images.

I have the same, identical, Images of the system partition on both HDDs, created one after the other to the target partition on each HDD.

Each operation of Creating, doing Integrity Checks and Restoring takes approximately 2.5-3 times longer on the 7K250.

I would expect reading from one partition and writing to a different partition on the same HDD takes longer compared to reading from one HDD and writing to a different HDD but since an Integrity Check is a read operation only, there shouldn´t be that much of a difference.

The transfer rates, according to Ghost during an Integrity Check is, for the 7K250 142 MB/s and for the 60GXP 397 MB/s.

I promptly did a benchmark in AIDA32 with the following results:

Quick Linear Read - 7K250 = 58.3-28.1 MB/s (46.8 avg)

QLR - 60GXP = 38.8-18.5 MB/s (31.2 avg)

Random Read - 7K250 = 57.6-27.9 MB/s (46.0 avg)

RR - 60GXP = 35.2-17.8 MB/s (30.7 avg)

Buffered Read - 7K250 = max 73.9 MB/s (73.7 avg)

BR - 60GXP = max 74.4 MB/s (74.2 avg)

(min readings omitted, low dips due to unknown interference)

Average Access - 7K250 = 11.8-13.4 ms (12.6 avg)

AA - 60GXP = 12.2-13.8 ms (13.0 avg)

The 7K250 should outperform the 60GXP and it does in the AIDA32 benchmark but why the very low performance when running Ghost?

Thanks for Your time,

Christer

Share this post


Link to post
Share on other sites

let me make sure I understand that correctly:

7k250 has a "system" partition on it...

you go through a process of 1) creating ghost image of the system partition and place one copy on the 7k250 and another copy on the 60GXP.

Then you 2 do an integrety check

Then you 3)restore the files [to the 7k250].

did i get that?

I would imagine this would be because you are having to do more seeks on the 7k250 since you are 1) having to write the image while reading info from the same drive which is increasing the number of seeks, 2) having to read 2 different parts of the same drive that are in different locations, and 3) restoring again involves reading/writing to different locations on the same drive causing increased seeking.

The fastest operation would be the one that keeps everything sequential. For example, ghosting one drive to the other or restoring an image from one drive to another.

Share this post


Link to post
Share on other sites

Hi blakerwry,

thanks for Your response!

I have a parallel thread going at Radified. It was the natural place to start since the issue is Ghost related. I came here to see if the storage gurus had some experience to share with me.

My HDD setup is:

Primary Master - 7K250:

C: System Partition with WinXP and programs (NTFS)

D: Data Partition (NTFS)

E: dedicated to Ghost Images (FAT32)

Primary Slave - 60GXP:

F: BackUp Partition (NTFS)

G: dedicated to Ghost Images (FAT32)

The reason for two partitions for Ghost Images is that, even though the 60GXP has soldiered on flawlessly for three years, it´s called a DeathStar and some people are suspicious against the 7K250 as well. The general idea is that both won´t fail simultaneously ...... :unsure: ...... !

Since I started this thread, I have created the same Image to both HDDs. I have checked the Integrity and both passed. I have restored C: using both Images, just to time the operation.

In all cases the operations carried out on the 7K250 was 2.5-3 times slower than on the 60GXP.

I would expect a read-and-write operation from one partition to another on the same HDD to be slightly slower than reading from one HDD and writing to another HDD. I would, however, not expect a read only operation, like an Integrity Check, to differ in speed.

I have checked both drives using the IBM Drive Fitness Test and the IBM Feature Tool. Both HDDs are configured correctly (identically). In addition to that I have used AIDA32 to benchmark and the results confirm that they are both working properly.

I don´t think that it is a HDD configuration issue since the transfer rates are much below ATA-100 rates. An example according to Ghost readings during an Integrity Check:

CD-R/RW operating in Multi-Word DMA Mode 2 - 102 MB/minute (1.7 MB/second)

DVD-ROM operating in Ultra DMA Mode 2 (ATA33) - 107 MB/minute (1.8 MB/second).

7K250 operating in Ultra DMA Mode 5 (ATA100) - 143 MB/minute (2.4 MB/second).

60GXP operating in Ultra DMA Mode 5 (ATA100) - 397 MB/minute (6.6 MB/second).

I mean, not even the fastest rate gets near Multi-Word DMA Mode 2 (Ultra DMA Mode 0) which is 16.7 MB/second!

It is possible that I´m missing the obvious ...... :blink: ...... but if I´m not, I suspect that the 7K250 is misidentified by Ghost, which obviously is running the show.

I would be glad to be proven wrong, though!

Christer

(If it has any bearing on this issue, the processor is an AMD Athlon Thunderbird 1GHz/266MHz and 512 MB SDRAM PC-133.)

Share this post


Link to post
Share on other sites

DOS doesnt really utilize UDMA or any native caching unfortunately, so you're stuck with the slower speeds no matter what the drive.

You are missing the obvious. It only makes sense to me that the 7k250 would be slower for this operation..

To give you an example to help illustrate why this is take the process of copying a term paper.

Method1:

A single person has the term paper on his/her left and then copies word for word from the original paper to a new paper on the right.

Method2:

Two people work together, one person reads the paper while the other writes the paper.

The second method will go as fast as the person writing the paper, the 1st method should be slower because the person has to read AND regergitate the information and can only work in small chunks (one thing at a time instead of both simultaneously ).

This is the case for writing, but it is also the case for reading as well. If you wanted to error check the copies and compare them to eachother it would work best if you had two people working together vs one.

Share this post


Link to post
Share on other sites

I could almost make a bet that You´re a teacher of some kind ...... :) ...... You find good analogies!

Your explanation will "hold water" if the Integrity Check actually is a read and simulated write operation in order to check what would have been written at the other end.

Two things bother me, though:

1) With todays seek times, it should be a slight increase in time, not 2.5-3 times longer.

2) When the 60GXP was the only HDD in my system, it had the partitions that the 7K250 has now. The software installation was exactly the same and I distinctly remember that when doing Integrity Checks on the Images burned to CDs, it took forever compared to the Integrity Checks on the 60GXP. (Compare my figures for the rates of the opticals and the 60GXP and then with the 7K250.)

Christer

Share this post


Link to post
Share on other sites

I have created and checked another Image which shoots holes in Your read-write theory.

A reminder of my partitions:

7K250 - C: - D: - E:

60GXP - F: - G:

This time I created an Image of F: to E: as well as G: with the results below:

Target partition E: - create in 10min 41sec at 136 MB/min - check in 9min 59sec at 145 MB/min

Target partition G: - create in 4min 14sec at 342 MB/min - check in 3min 43sec at 390 MB/min

The different Image gave slightly different rates but it definitely points towards either some unknown (to me at least) feature of the 7K250 or some bug in Ghost.

Christer

Share this post


Link to post
Share on other sites

Is someone else using Ghost in combination with the HDS722512VLAT80 (or any other Hitachi 7K250?

There must be someone and I would very much appreciate any comments on how it performs with Ghost!

Thanks,

Christer

Share this post


Link to post
Share on other sites
Is someone else using Ghost in combination with the HDS722512VLAT80 (or any other Hitachi 7K250?

There must be someone and I would very much appreciate any comments on how it performs with Ghost!

Thanks,

Christer

my 7k250´s behave Ok in DOS version of Norton Ghost (2x160GB PATA 8MB cache) and i have both on same channel and backup was from one 7k250 to second 7k250 :huh:

Share this post


Link to post
Share on other sites

Hi KuKla,

thanks for Your response!

What transfer rates do You have when checking an Image? Would be nice (?) to compare with my average of 142 MB/min.

To compare our computers, on which motherboard and processor is Yours based? Mine is based on EPOX 8KTA3 and 1GHz/266MHz Athlon Thunderbird.

Best regards,

Christer

Share this post


Link to post
Share on other sites
Hi KuKla,

thanks for Your response!

What transfer rates do You have when checking an Image? Would be nice (?) to compare with my average of 142 MB/min.

To compare our computers, on which motherboard and processor is Yours based? Mine is based on EPOX 8KTA3 and 1GHz/266MHz Athlon Thunderbird.

Best regards,

Christer

my avg was about 750mb/min (AFAIK)...

my PC configuration is in signature.....epox 8rda+ (nForce2), barton 2500+, 512megs of DDR...

Share this post


Link to post
Share on other sites

Well, I obviously need to set up an appointment with my optician!

Thanks for the information but I forgot to ask which firmware version?

Best regards,

Christer

Share this post


Link to post
Share on other sites

Hi sgrossklass,

thanks for the link!

I´m not that proficient when it comes to computers but I have saved the drivers to my HDD.

Maybe, one day, I´ll learn how to use them!

Christer

Share this post


Link to post
Share on other sites

There IS usually a big difference when copying from one partition to another on the same drive compared to copying from one disk to another. And yes that speed difference is around 2-3 times according to my experiences.

It's very easy for you to do the test yourself. Just create a 1GB folder of files with suitable sizes on your C or D drive and try copying to your E drive. Then try copying from your C or D drive to your G Drive. If you measure the time it takes to finish this operation I bet you will see that it takes at least twice as long time to copy between the 2 partitions on the same drive.

blakerwry already tried to point this out to you.

Share this post


Link to post
Share on other sites

well orchid, even with that fact it seems the 60gxp simply performed better after some basic testing... must be something about his drives or his setup.

Share this post


Link to post
Share on other sites

Oh yeah sorry I missed that one.

But I think it would still be an interesting test to do. We have seen that it performs bad when creating the ghots image - now let's see if the same is true during file copying.

Share this post


Link to post
Share on other sites
With todays seek times, it should be a slight increase in time, not 2.5-3 times longer.

That is where the misunderstanding is, because it is not the seek time that causes the delay but the rotational latencies plus the seek times. Chances are that you are missing the target LBA on every access regardless of whether you are doing reads from the "Ghost" partition or writes to the target partition and that will cause huge delays.

FYI, I have timed about 30 different drives over the past 7 years with respect to file copy from one partition to another as opposed to file copy from one drive to another or even file copy via Ethernet and the average performance hit is in the order of 3 - 5 x execution time if the files are copied within the same drive from one partition to the other.

Keep in mind also that you are copying from ID to OD in your case which runs against any track skew optimization of the platter and poses an additional obstacle.

Share this post


Link to post
Share on other sites

maybe it would also be nice to point out that seek times havent improved in years... and infact the trend seems to be that seek is getting worse if anything.

@ unregistered

Keep in mind also that you are copying from ID to OD in your case which runs against any track skew optimization of the platter and poses an additional obstacle.

what do you mean? ID? OD?

Share this post


Link to post
Share on other sites

Hi guys!

orchid5,

It's very easy for you to do the test yourself. Just create a 1GB folder of files with suitable sizes on your C or D drive and try copying to your E drive. Then try copying from your C or D drive to your G Drive.

I did almost as You suggested, I had a folder on D: holding 611MB of install files - Windows service packs and updates and other setup files:

Copy from D: (7K250) to E: (7K250) took 67 sec => 9.1 MB/sec

Copy from D: (7K250) to F: (60GXP) took 39 sec => 15.7 MB/sec

Copy from F: (60GXP) to G: (60GXP) took 114 sec => 5.4 MB/sec

I´m sure this proves something but to me it´s confusing. What´s all the mumbo-jumbo about the 7K250 having a maximum internal data rate of 757 Mb/sec (~94 MB/sec) and the 60GXP 494 Mb/sec (~62 MB/sec)?

The difference between D: => E: and F: => G: reflect the difference between the drives but approximately ten (10) times lower than on paper. Do seek time and latency really have this impact on performance?

blakerwry already tried to point this out to you.

Yes, he did but we obviously talked past eachother. My comment was:

Each operation of Creating, doing Integrity Checks and Restoring takes approximately 2.5-3 times longer on the 7K250.

I would expect reading from one partition and writing to a different partition on the same HDD takes longer compared to reading from one HDD and writing to a different HDD but since an Integrity Check is a read operation only, there shouldn´t be that much of a difference.

Attempt to clarify:

As soon as the target partition is on the 7K250, every operation is 2.5-3 times slower compared to when the target partition is on the 60GXP. The location of the source partition doesn´t make a difference.

blakerwry,

... must be something about his drives or his setup.

I have tried any and all combinations, including disconnecting the optical drives and moving the 60GXP to Secondary Master to have one drive alone on each channel. The results are the same ...... :blink: ...... !

unregistered,

... the average performance hit is in the order of 3 - 5 x execution time if the files are copied within the same drive from one partition to the other.

The results from my "single simple test" above are different. The performance hit was "only" a factor of 1.7.

I have created Images of C: (7K250) to E: (7K250) and G: (60GXP)

I have created Images of F: (60GXP) to E: (7K250) and G: (60GXP)

It doesn´t matter where the source partition is located, it is the target partition that is the deciding factor. If it is on the 7K250, each and every operation (create, check, restore) is 2.5-3 times slower.

I created an Image of F: (60GXP) to E: (7K250) at 136 MB/min and it was checked at 145 MB/min.

I created the same Image of F: (60GXP) to G: (60GXP) at 342 MB/min and it was checked at 390 MB/min.

This shoots holes in some theories and indicates that it is neither a read/write problem nor has it any connection to seek times and latency.

I´m inclined to believe that there is something fishy about the 7K250. The question is if it is only mine or every 7K250?

Christer

Share this post


Link to post
Share on other sites

Well you have now noticed that the 7K250 is very fast at copying from one partition to another (something I noticed myself).

But you missed maybe the most important copy test.

We saw that creating an image from 60GXP --> 60GXP was fast.

We also saw that creating an image from 60GXP --> 7K250 was slow.

So now let's see the 'same' file copy test as you already did but this time from 60GXP --> 7K250 to see if that is slow as well.

That might point us to the problem.

Share this post


Link to post
Share on other sites

Yes orchid5, You´re right.

I meant to do it but couldn´t use the same file in the other direction and I forgot ...... <_< ..... lame excuse, eeh!

I copied a backup of the previous folder (615 MB, not the exact same contents and size, something has been removed or changed since I created the backup) from F: (60GXP) to D: (7K250) and it took 49 sec => 12.6 MB/sec which is 80% of the speed in the other direction.

There is absolutely no logic to the rates in Ghost when the target partition is on the 7K250 and in my exchange of e-mails with Hitachi, it seems like their hair is turning grey!

Christer

By the way, neither drive makes a sound while performing the tests which is nice.

When, on a friends computer, I copied some 30 GB from one Seagate 7200.7 to another, I noticed that they were very loud.

On the whole, I´m satisfied with the 7K250 but the Ghost issue bugs me.

Share this post


Link to post
Share on other sites
created an Image of F: (60GXP) to E: (7K250) at 136 MB/min and it was checked at 145 MB/min.

I created the same Image of F: (60GXP) to G: (60GXP) at 342 MB/min and it was checked at 390 MB/min.

This shoots holes in some theories and indicates that it is neither a read/write problem nor has it any connection to seek times and latency.

Maybe I missed it but what are the metrics you are using? I mean, how are you deriving those data? The numbers I quoted were obtained using a stopwatch.

Share this post


Link to post
Share on other sites
Maybe I missed it but what are the metrics you are using? I mean, how are you deriving those data? The numbers I quoted were obtained using a stopwatch.

During the operation, Ghost displays the following information on the screen:

Actual data rate, MB processed, MB still to process, time lapsed and time to complete.

When finished, Ghost displays an average data rate, total processed MB and total time.

I checked the Ghost timing with my chronograph and divided MB with time and it gets it right.

My own tests were timed using a chronograph.

Christer

(Chronograph = wrist watch with integrated stop watch)

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now