Raptors in RAID-1

Ryom

Ryom

[H]ard|Gawd
Joined
Oct 11, 2006
Messages
1,859
Anyone have a pair of Raptors in a RAID-1 array that can post some HDTune benchies? Preferably the 36GB ADFD Raptors, but any pair of Raptors will do.

I just want to compare is all and I'm not finding any useful benchmarks via google.

HDTune_Benchmark_NVIDIA_MIRROR_34.47G.png
 
im no RAID buff, but wouldnt Raid1 yield the exact same performance as a single drive (perhaps a tad less due to controller overhead)?

AFAIK the only use for Raid 1 is in preventing data loss when a single drive fails, at the cost of the capacity

did you mean Raid 0 (striping) ?
 
No, I do mean RAID-1. You should be able to get split reads from RAID-1 which could increase read performance and seek performance. Write performance won't see any benefit. I was mainly after redundancy, so any performance gain is icing on the cake. If anyone has the Raptors in a RAID-0 array I wouldn't mind seeing that either... although there are already a few RAID-0 benchies of the Raptors to be found online.

It seems the mirror array definitely has improved read performance vs a single drive, but not quite on par with a pair in a stripe.
 
Redundancy? Are you seeking this solution as a desktop/gaming alternative or will this be server type duties?
 
graph looks very similar to me when i had four raptors in raid 1
 
Four in RAID-1! That's certainly playing it safe :)

In theory your access times should be noticeably lower on average (vs single disk) for that kind of setup if your controller does split seeks. Got any stats saved?
 
Ryom said:
No, I do mean RAID-1. You should be able to get split reads from RAID-1 which could increase read performance and seek performance. Write performance won't see any benefit. I was mainly after redundancy, so any performance gain is icing on the cake. If anyone has the Raptors in a RAID-0 array I wouldn't mind seeing that either... although there are already a few RAID-0 benchies of the Raptors to be found online.

It seems the mirror array definitely has improved read performance vs a single drive, but not quite on par with a pair in a stripe.
Click to expand...

hmm, i had thought about read-splitting ability in raid-1 before, but i didnt know this was implemented already, thats pretty cool

as for your rationale behind the Raid-1 raptors, i understand your point, but is the risk of OS drive failure really that significant? my view on the situation is that an OS drive failure is so unlikely that it isnt worth running raid 1 for to prevent, since i have no time or otherwise critical data on my system (i do off course run backups periodically to a usb disk)

im just wondering if your data is critical enough to warrant this setup, not to mention the fact that hard drive failure is quite rare (if it is not, it might be indicative of another problem)
 
My data isn't critical per se, but a hundred bucks isn't an overly extravagant investment either.. what I spent for a second drive pays for itself the first time the redundancy saves me from restoring from backups. Thats a good investment in my mind. Not to mention it seems I may be getting a fairly respectable performance gain out of it.

And for the record, I've never had a disk failure on my personal machines :)
 
Ryom said:
My data isn't critical per se, but a hundred bucks isn't an overly extravagant investment either.. what I spent for a second drive pays for itself the first time the redundancy saves me from restoring from backups. Thats a good investment in my mind. Not to mention it seems I may be getting a fairly respectable performance gain out of it.

And for the record, I've never had a disk failure on my personal machines :)
Click to expand...

hmm, the performance increase has me interested, does stuff really load faster from the array?

and i suppose you do have a point, when you are a working man, investing 100 dollars to prevent having to spend hours on restoring a critical tool is a non decission, from a bussiness standpoint this makes a huge amount of sense *mumbles something about his current empoyer using laptops for workstations*
 
DrFreeze said:
hmm, the performance increase has me interested, does stuff really load faster from the array?
Click to expand...

I need to pull some more benchies of single raptors up for comparison. The reads seem a bit high for a single drive since they are kicking my 7200.10 750GB drive around. Seek times would be worse, but I was under the impression the latest 7200rpm drives had similar throughput to the raptors.
 
Ryom said:
No, I do mean RAID-1. You should be able to get split reads from RAID-1 which could increase read performance and seek performance. Write performance won't see any benefit. I was mainly after redundancy, so any performance gain is icing on the cake.
Click to expand...
There are very few controllers that are `advanced' enough to provide any benefits from RAID-1. I think I saw some info on Anandtech that showed a slight improvement in seek time for two or three controllers. Maybe they have some IOmeter results that can show the improved read performance with higher queue lengths.

Ryom said:
http://www.overdrivepc.com/forum/showthread.php?t=13036&page=1&pp=15

If you are interested that link will take you to a summary of my disk system setup and the reasoning behind it.
that link said:
chose to RAID-1 the Raptors because I intensely dislike the processes and time loss of reinstalling or restoring an OS (operating system) and its associated programs to a replacement disk. Who at home is going to keep a spare harddrive around simply to swap in in case your current one fails? Not many. So the restore time becomes (time to buy, ship, install drive, and restore data) or (time to find older drive in basement + restore, then time to buy, ship, install drive, and restore data to the preferred replacement drive) Yuck. I don't understand that mentality. So I buy both drives ahead of time and get the boosted read performance of a RAID-1, with downtime due to disk failure reduced to zero and a restore time of the short time it takes to rebuild the array (excluding of course the double disk failure... but in those instances the cause of the dual disk failure was probably a system killer anyway ) I say RAID-1 should be the default setup for the OS drive!
Click to expand...
Click to expand...

The problem with your reasoning is that you assume that a hardware failure is what leads to most unbootable OS installs. I have not statistical data to back me up, but if looking around this forum is any indicator, most data loss and corruption occurs due to user error. RAID-1 (or any RAID-n, n > 0) does not provide any type of protection against user error. Delete or modify an important windows file on one drive, and you will have deleted or modified it on the other as well.

RAID-1 does one thing only: it increases the expected uptime of your hardware.
 
Yes, I'm fully aware that RAID-1 is NOT A BACKUP! I have an external drive that I do incremental backups to. The array is there so that I don't have to restore from backup should a drive fail which would be inconvenient since I don't do nightly backups. Because IF that SINGLE drive fails I would be FORCED to restore from backups!

So what exactly are the other things wrong with the thread? I've noticed that my reads seem higher than what seems normal and I'm trying to verify by collecting data and I've made some suppositions meanwhile as to why the reads may be high.


The problem with your reasoning is that you assume that a hardware failure is what leads to most unbootable OS installs. I have not statistical data to back me up, but if looking around this forum is any indicator, most data loss and corruption occurs due to user error. RAID-1 (or any RAID-n, n > 0) does not provide any type of protection against user error. Delete or modify an important windows file on one drive, and you will have deleted or modified it on the other as well.
Click to expand...

And the problem with your reasoning is that you assume I don't already know all of that. I do. The array is there in case of drive failure. The backup drive is there in case of any other cause. You read my OPC post, but apparently skimmed over or ignored all the important bits.
 
Storage Review's tested specs for a 36GB Raptor:
Max Transfer: 63MB/s
Min Transfer: 45MB/s

My 36GB Raptor RAID-1 array's tested specs:
Max Transfer: 84.5MB/s
Min Transfer: 53.5MB/s



The mirrored array has a max read speed 21.5MB/s higher than what was measured for a 36GB single drive and a min read speed 8.5MB/s higher. The array is 33% faster at max transfer and 20% faster at min transfer than the same single drive.

The array's max transfer also beats the 74GB Raptors max transfer by 12.5MB/s and ties the minimum transfer.
Once pitted against the largest and most current Raptor, the RaptorX 150GB, the 36GB array almost ties it. The array's max transfer is 3.5MB/s lower, and its min transfer is 6.5MB/s lower.

So drizzt81 and smizack, why are my 36GB Raptors performing well above a single 36GB 2nd gen Raptor and nearly on par with the 150GB RaptorX?
 
Ryom said:
Storage Review's tested specs for a 36GB Raptor:
Max Transfer: 63MB/s
Min Transfer: 45MB/s

My 36GB Raptor RAID-1 array's tested specs:
Max Transfer: 84.5MB/s
Min Transfer: 53.5MB/s
Click to expand...
It would have helped had you posted some links.
Here is a comparison on SR between the WD360GD Raptor (first gen, which you quoted) and the WD1500ADFD Raptor (lastest gen). According to various sources on the internet, the new WD360ADFD Raptor performs very much like its larger brother.

Looking at the number for the WD1500ADFD, you will note the following:
Max Transfer: 88.3 MB/sec
Min Transfer: 60.2 MB/sec
Click to expand...

Since you did not specify which Raptor you have -apart from size-, nor did you say which controller you are using. Since these two variables can change the outcome quite a bit it is almost impossible to tell you why your Raptors are performing well above the numbers that SR has for the WD360GD.

Let me go out on a limb here and take a look at the first post you made: you are using WD360ADFD Raptors, not the first generation piece-of-crap. If you search this board thoroughly, you will see that quite a few members do not recommend the first generation Raptor at all anymore, since even the smallest latest gen drive is better than them. The WD360GD was the first in the line of 10k rpm drives from WD and its performance was well surpassed by the 74GB version that WD released afterwards.

Does this answer your question?
 
Sure, but you didn't say that in the first part of your post, and in the second part you reiterated back to me what I had previously stated in my first post on OPC as though I hadn't stated it. That bound to make anyone a little irked.

If you've got some HDTune benchies of a single 36GB AFDF you'd like to share, I'd love to see them. I'm not finding any reliable sites online for comparison. Hence, this thread. If you don't think that RAID-1 does anything and want to weigh in, then help me verify it one way or the other.
 
  • THG.com has tested the ARC1220 in RAID-0,1,0+1
    • Access time in H2wBench is a bit lower in R-1 than in R-0 or R-10.
    • STR in R-1 is a little morethan 1/2 of the 2-disk R-0 array for read operations, less then 1/2 for write operations.
  • Tweakers.net has this comparison of a single WD740ADFD on a LSI MegaRAID card vs. a two-disk RAID-1 array on the same card. As you can see the Access time is about 9% faster on the RAID-1 array. The difference in ATTO STR is around 1%, which I would say is most likely within the margin of error, and Winbench delivers the same results for both configurations.
  • Tweakers.net also has a comparison of nine SATA RAID controllers. According to the chart titled Winbench 99 v2.0 - Average access time (ms) some controllers improve the average access time when comparing RAID-1 to a single drive, while others seem to fare better with a single drive.
    The charts below show the STR (which is what HDTune tests if I am not mistaken) and you will notice that the results are -once again- mixed depending on the controller.

I am sorry for misleading you into visiting anandtech in my earlier post. I confused them with Tomshardware despite both sites being rather different.

My statement from before stands as I had made it:

There are very few controllers that are `advanced' enough to provide any benefits from RAID-1. I think I saw some info on [website] that showed a slight improvement in seek time for two or three controllers. Maybe they have some IOmeter results that can show the improved read performance with higher queue lengths.
Click to expand...

My last sentence appears to be somewhat true as the charts show a generally upwards trend when going from a single disk to RAID-1.

ps.: If you do not consider Tweakers.net a reliable site, I am sorry for wasting your time.
 
I tried to drop this subject, but since you insistend on bringing it back up:
Ryom said:
Sure, but you didn't say that in the first part of your post, and in the second part you reiterated back to me what I had previously stated in my first post on OPC as though I hadn't stated it. That bound to make anyone a little irked.
Click to expand...

I assumed (yes bad part on my side) that you wanted a fair comparison between the performance of a RAID-1 array and a single disk. This requires that either the comparsion is done using the same[1] drives or that some form of scaling is used to make the numbers comparable. If we play the mix-and-match game, I can surely find or prepare plenty of benchmarks where a RAID-1 system fares far worse to a single disk, even in HDTune. I have a couple of IBM harddrives with less than 20GB and a current generation Raptor. I do not need a crystal ball to see that the Raptor will beat the RAID-1 array in almost any harddrive related benchmark.

Ryom said:
If you don't think that RAID-1 does anything and want to weigh in, then help me verify it one way or the other.
Click to expand...
I said that the performance advantage is small and controller dependent. Additionally, it appears that some applications (IOmeter with high queue depths) may benefit more than others.

[1] even a variation in the firmware can have a `large' impact on the performance: as suggested here.
 
I don't mind being wrong, I just want the people that disagree to, at the very least, attempt to back it up... which you've now done. Thank you :)
 
You must log in or register to reply here.
Back
Top