WD 1000 JB (8 mega cache) benchmarks

[davehole]

Originally posted by chainbolt
chainbolt.com: Winbench 99 (2.0) score for 2 raided 1000JB: 62,000

lol, chainbolt.com

why does this not exist???

davehole

[chainbolt]

ok, oh my time ................... what a waste of my life!


here is the Winbench 99 disk inspection STR diagram for 2 raided 1000JB disks optimized by chainbolt.com




and this here is a single 1000JB disk optimized by auenf.com

http://www9.brinkster.com/sportschoo...d1000jbstr.jpg

Comparing ONLY the first 20GB of the 2 diagrams, I hope everybody can see the difference. Auenf: it's certainly more than 20% as you said! If you compare my 20GB with the first 20GB of your diagram, I would say it's 90 MB/sec against 45 MB/sec, that is a difference of 100%. And in the Winbench 99 High-end Disk marks the difference is even bigger:

1 single 1000JB at storagereview.com: about 23,000
2 raided 1000JB at chainbolt.com: about 62,500

[auenf]

ok, the winbench99 STR seems to look fine, but your very first HDTach benchmark doesnt look as good, ie 55MB/s v my 45MB/s, which is what i confused about, and where the 20% came into it

im just concerned that checking that your HDTach str is a little low

Enf...

[chainbolt]

Originally posted by auenf
ok, the winbench99 STR seems to look fine, but your very first HDTach benchmark doesnt look as good, ie 55MB/s v my 45MB/s, which is what i confused about, and where the 20% came into it

im just concerned that checking that your HDTach str is a little low

Enf...

I see the point, you are comparing HDatch STR. Actually, I value HDtach more than other STR benchs. Would it be possible that you post your Hdatch screenshot for a single 1000JB?

[chainbolt]

ok, I see it now, you are talking about this HDtach graph for 1 1000JB in comparison with mine for 2 raided 1000JB:

1 single 1000JB Western Digital (auenf)

http://www9.brinkster.com/sportschoo...wd1000jbhd.jpg

2 raided 1000JB Western Digital (chainbolt.com)




May I explain this?

a) my RAID array is 200GB. I get an average of 46 MB/sec.
b) your single disk is 100GB, and you get an avrage of 37 MB/sec

Now, as already mentioned before: the transfer rate is dropping the more you approach the inner portions of the disk, because they are rotating slower than the outer portions. This effect is naturally much higher if you calculate the average MB/sec for a 200GB disk than for a 100 GB disk. This is the reason why ONLY a comparison of same size disks is usful when talking about STR.

Arranging a 100 GB array (comparable with your single 1000JB disk) would yield an average of 50MB/sec as deductable from my HDatch screenshot. That would be around 40% more than what you get with a single disk, and pretty much in line with the expecations of RAID 0 versus single disk.

But even now: 100GB single versus 200GB raided: it's already 25%, that's a lot I think, and the REAL impact on applications is MUCH higher than the STR difference is indcating as we can see in the posted Winbench 99 High-end diskmarks for single and raided 1000JB disks:

single 1000JB disk: WB99 High-end disk score: 23,000
raided 1000JB: WB99 High-end disk score: 62,500


For all who are reading this: don't get confused, both transfer rates for the single and the raided 1000JB Western Digitals are outstanding results.

[KyocEr@~*]

chainy, can u give us a link to how to tweak our hard drives... don't have a raid array yet, but going to very soon!

thanks man, thanks.

[chainbolt]

Originally posted by KyocEr@~*
chainy, can u give us a link to how to tweak our hard drives... don't have a raid array yet, but going to very soon!

thanks man, thanks.

only if you explain the meaning of this to me:

[KyocEr@~*]

http://userfriendly.org/static

Only after you read through every single comic strip from a few years back on that website! =)

It's "Dust Puppy" - a character there.

[chainbolt]

Originally posted by KyocEr@~*
chainy, can u give us a link to how to tweak our hard drives... don't have a raid array yet, but going to very soon!

thanks man, thanks.

ok, have a look at this thread I started at www.amdmb.com a few weeks ago. It's a very lonooooooooong read now. But contains all information about stripe/cluster optmization currently available. You will see there some WB99 scores even better than what I have posted here

Result:

*for the Maxtor D740X a stripe/cluster combo of
32K/32K seems to be the best: look for the scores and comments of Bob.D)

* for the 1000/1200JB a stripe/cluster size of 16K/16K seems to be best: look of the scores and comments of ACE-La Rue.

http://www.athlonmb.com/vb/showthrea...hreadid=109490

[Step]

Thanks, might have to try 32k/32k for my Maxtors.

[DigitalMonk]

Thanks for the info chainy. I'll keep it in mind when I reformat in the next couple weeks.

I'm looking into a new CPU, hopefully in the next couple weeks, but maybe not till august.

[chainbolt]

By the way, I hear that the IBM 120GXP are also great performers, close to the 8 Mega cache Western Digitals, close, and I think there is still this reliability issue with the GXP technology.

[xtreme2k]

I have heard one of my friend raid0 with 2 IBM 120GXP getting about 52000 sandra hdd mark

I cannot get him to do a screenshot just that but I will try to.

[chainbolt]

Originally posted by Step
Thanks, might have to try 32k/32k for my Maxtors.

ok, here is the way how to format NTFS in 32/32 (kudos to GOD):


The easiest way is with 3 disks, of course. Install the OS to the 3rd disk, parititon and format the RAID 0 array as you wish, and then install the OS on the array. But most ppl don't have 3 disks. If you want to use NTFS you cannot format from a DOS disk.

This here is the way to out-smart the W2K or XP installation routine, which always, by default, formats NTFS in 4k clusters. As a result you get your W2k/XP OS installed on a RAID 0 array with a NTFS file cluster size as you desire (and not only the default 4K).

1) After connecting the 2 disks to the RAID interface, power up, go into the RAID controller BIOS and set up a RAID 0 array with the desired stripe size, be it for the Highpoint controller (as we have it on Abit or Epox boards) 16K, 32K, or 64K, or in case of a Promise controller (Asus for example) 4K, 8K, 16K, 32K, 64K, 128K, 256K, 512K, and 1024K.

(2) Then you reboot and start installing W2K/XP. After properly loading the RAID controller driver from a prepared floppy by hitting "F6" at the beginning of the installation routine you arrive at the blue W2K/XP installation screen. It will ask you where to install W2k/XP. At that moment you have the choice to set up and format partitions. The problem is that you can only format in the default size of 4 K, which is not what we want. But this is what we do: you partition and format at least 2 drives: c:/ and d:/. The c:/ drive will later be used as the "real" installation drive for W2k/XP. The d:/ drive is where you install XP now as a temporary solution. Choose whatever size you wish for c:/ (it can later be changed) and give the temporary installation drive d:/ the minimum size you need for a proper XP installation, which is 2GB.

(3) Then you install W2k/XP on the d:/ drive. After the installation is completed, run W2k/XP and go to "My Computer", right click "Manage", call for "Disk Management" and RE-PARTITION and REFORMAT the empty c:/ drive (which was previously formatted with the default 4K cluster size) with the desired NTFS cluster size, be it 8K, 16K, 32K, or as maximum 64K. Make SURE to format c:/ as PRIMARY partition and set it ACTIVE afterwards. Then you reboot and re-peat the W2K/XP installation process. But this time, when arriving at the partition screen, you will tell W2K/XP to delete the d:/ drive with the temporary W2k/XP installation. Then you install W2k/XP fresh on the already existing c:/ drive, which you previously formatted in the desired cluster size.

This works 100%. It is, as far as I know, the only way for ppl. with only 2 disks to format a RAID 0 array in NTFS in another than the default 4k cluster size. And 4K is NO good for RAID 0 performance, as everybody who has read through this thread knows by now

[auenf]

Originally posted by chainbolt
ok, I see it now, you are talking about this HDtach graph for 1 1000JB in comparison with mine for 2 raided 1000JB:

1 single 1000JB Western Digital (auenf)

http://www9.brinkster.com/sportschoo...wd1000jbhd.jpg

2 raided 1000JB Western Digital (chainbolt.com)




May I explain this?

a) my RAID array is 200GB. I get an average of 46 MB/sec.
b) your single disk is 100GB, and you get an avrage of 37 MB/sec

Now, as already mentioned before: the transfer rate is dropping the more you approach the inner portions of the disk, because they are rotating slower than the outer portions. This effect is naturally much higher if you calculate the average MB/sec for a 200GB disk than for a 100 GB disk. This is the reason why ONLY a comparison of same size disks is usful when talking about STR.

Arranging a 100 GB array (comparable with your single 1000JB disk) would yield an average of 50MB/sec as deductable from my HDatch screenshot. That would be around 40% more than what you get with a single disk, and pretty much in line with the expecations of RAID 0 versus single disk.

But even now: 100GB single versus 200GB raided: it's already 25%, that's a lot I think, and the REAL impact on applications is MUCH higher than the STR difference is indcating as we can see in the posted Winbench 99 High-end diskmarks for single and raided 1000JB disks:

single 1000JB disk: WB99 High-end disk score: 23,000
raided 1000JB: WB99 High-end disk score: 62,500


For all who are reading this: don't get confused, both transfer rates for the single and the raided 1000JB Western Digitals are outstanding results.

ok, i was under the impression that striping the two discs would give a bigger than 25% increase, which is why i queried it in the first place.

Enf...