fdisking, partitions and RAID - oh my!

[gradius]

I had an idea in my head, but I wasn't sure on how to explain it, so I fired up Photoshop to illustate my question. Is this possible?

Take 2 hard drives:

http://www.virtualtrends.com/gradius/1drive.gif

Partition the 13GB:

http://www.virtualtrends.com/gradius/2drive.gif

Then use a RAID card to make the two 8GB partitions one drive C:\ leaving the 5GB partition as the drive D:\:

http://www.virtualtrends.com/gradius/3drive.gif

Since the drives are of unequal size, with the partitioning, I wouldn't lose any space while using RAID.

[This message has been edited by gradius (edited 07-09-2000).]

[hd581]

I've set up a Striped Set (RAID-0) in NT Server 4 with unequal hard drives and it selects the largest matching portion of each hard drive as you illustrated. It's possible through software RAID so I imagine hardware RAID should have no trouble doing it.

BTW, I just had to comment on how remarkably clear your illustrations made your point.

[This message has been edited by hd581 (edited 07-09-2000).]

[gradius]

thank hd581. I figured I'd spend a few minutes showing us visual learners what I meant, than to try and spend pages explaining an easy-to-see concept.

Ok, you metioned "software" RAID? I didn't know such a thing existed. How does it differ from using a RAID card?

[This message has been edited by gradius (edited 07-09-2000).]

[Stan]

Hi

Software RAID means that RAID arrays are created and maintained by the OS (NT for ex).
Therefore, software RAID needs to allocate itself some resources (from OS, CPU). There is an impact on general performance.

Hardware RAID (i.e. with a RAID card controller) is totally transparent. The OS "does not know" that there is a RAID card. It fools the OS and presents the array as one whole disk/partition.

With hardware RAID, disks have to be the same size.
If you want to implement hardware RAID with your 2 HD from your config, you would loose the remaining 5GB of the 2nd HD.

So, one advantage on soft RAID is that you can still use the remaining GB.

Hard RAID is better and offers better performance. It does not use OS resources (or CPU) as it has its own hardware controller.
Unfortunately, it's more expensive...

Hope it's clear enough... http://sysopt.earthweb.com/forum/smile.gif

Stan

[gradius]

actually, it's crystal clear Stan - thanks!

I'm really wanting to upgrade for the performance, so I might as well slap down some hard-earned and get myself a Promise RAID card. Of course, if I look around enough, I think there are still people selling modded ata-66 cards for cheap. I also have been looking into a 7200RPM drive. I hear the performance increase is pretty nice coming from a 5400RPM drive. Who knows? I guess I'll try to find the best price/performance ratio between the two.

[Stan]

Yep, FastTrak (or mod ULTRA) is the way to go.
I have a FastTrak with 4 IBM (7200rpm, 2Mb) configured in RAID0 (stripe).
I am flying... http://sysopt.earthweb.com/forum/smile.gif

Maybe you want to wait for the SuperTrak... (RAID5 IDE controller)

Stan


Go for 7200rpm disks, they are muuuuch faster than the good old 5400rpm


[This message has been edited by Stan (edited 07-10-2000).]

[hd581]

With hardware RAID, disks have to be the same size.I stand corrected. I wonder though: Can RAID cards other than fast track (DPT, DAC960, PM3334) handle different drive sizes?

[Mntsnow]

In all the hardware raid systems I have used can use different sized HD's BUT they will only use the capacity of the smaller HD. You will waste the extra space on the larger HD.

In gradius example in a HARDWARE raid set up he will lose 5 gigs. He could stripe the two togather and end up with a single 16 gig partition or multiple partitions that add up to a total of 16 gigs

[This message has been edited by Mntsnow (edited 07-10-2000).]

[Stan]

Sorry if it wasn't clear enough...

Like Mntsnow said: in hardware RAID, disks don't have to be the same size/model to be able to create an array.
The array will be based on the smaller sized HD and you would then loose the "extra" space of the bigger HD.

Stan


[This message has been edited by Stan (edited 07-11-2000).]

[entity]

Software RAID is amazingly fast as well, provided you have the right hardware to work with.

I have Win2000 Pro with 2 IBM 7200rpm UW-SCSI HDs, 9GB each, set up in software RAID-0 mode (both are "dynamic" disks).
The two disks are on a separate SCSI controller to avoid bottlenecks and I can retain the maximum flexibility.
The speed I can get is incredible.

Do not expect much higher figures for hardware RAID configs.

Hope to have helped you all.
Any question is welcome.

Bye,
EnTiTy

[Britewood]

Help me to understand why two SCSI controllers? The idea of SCSI is to allow multiple devices (drives) to communicate simultaneously without sharing bandwidth. Because it is designed for this and has such high transfer speeds I can't see how 2 hard rives on one SCSI controller would even come near maxing out the controller? Especially when most controllers can handle 7 or more drives. I'm sure you may have some performance boost but it seem like a waste of PCI slots for what tiny boost you do get, unless you plan on running several more drives? Maybe the boost you get is because your running Software RAID? Fascinating!

[Stan]

I think you found the answer yourself...

Ideal situation would be one controller per disk.
I makes even more sense in RAID 1 or 5.
That would be enhanced redundancy. If you loose a controller, you do not loose the entire array. You just just one HD.

Expensive solution though... these SCSI controllers are not the cheapest !!!

Stan

[Britewood]

Well if you are talking RAID 1 or 5 I can understand that. I currently run this configuration on my servers and have redundant RAID hardware controllers. But RAID 0? If you lose a controller or a drive… All Gone! All of it. The disadvantage of RAID 0 setup is that…

Not a "True" RAID because it is NOT fault-tolerant.

The failure of just one drive will result in all data in an array being lost

Should never be used in mission critical environments

This means that if he loses a SCSI controller or a Drive he will loose all data. Hence I ask again why 2 SCSI controllers? Or is it because he does get better performance as a result of running software RAID?

For some reason I almost feel that I am missing something?


MMMM!

Britewood

[Stan]

Yep, I agree with you...

That's the reason why I mentionned RAID1 and RAID5, not RAID0

Like you said there is no redundancy in a RAID0.
The only reason why you would use 2 (or more) controllers in a RAID0 array would be for increase of performance, nothing else...

At home, I have a RAID0 with 4 IBM. It runs extremely fast but if one disk goes... then I am in big trouble...
But why should I care, it is just a gaming PC ! http://sysopt.earthweb.com/forum/smile.gif

Stan

[This message has been edited by Stan (edited 07-20-2000).]

[Britewood]

On that note I understand! Gaming machine says it all!

Britewood

[This message has been edited by Britewood (edited 07-20-2000).]

[KillerBug]

Go extreme! Pick up 4 of those new 75gb IBMs, mod yourself an ultra, and stripe them, flying will be slow.

[Stan]

mmmm... 300GB of HD space... mmmmm....

[entity]

For Britewood and others:

Using two (or more) controllers is the only solution and I explain you why:

The typical transfer rate of each of my 7200rpm HD is about 20 MB/s.
My controller is UltraWide (I do not want to mess up with LVD termination troubles) and so has a bandwidth of 40 MB/s.
Two RAID-0 disks will almost saturate the whole the bandwidth.

I have also two Fujitsu 10.000rpm 36GB HDs (MAF3364LP) and they are even faster.
So did you realize where my need of multiple controllers comes from ?

Yes, RAID-0 is not fault-tolerant, but who cares ??? Performance is the key for me (and a good backup unit, of course...).

Bye now,
EnTiTy

[Stan]

I think that we all understood that you have 2 controllers for performance only... http://sysopt.earthweb.com/forum/smile.gif

Makes sense to me...

Stan

[entity]

All but Britewood it seems.

Anyway "repetita iuvat" Romans said.

Bye now,
EnTiTy

[hoods2]

Thanks guys,something new to play with.
But now I got to buy another scsi controller.
Is there that big of a gain with two controllers?
Thanks Dave

[entity]

If you read the topic history you will get the picture.

Calculate the theoretic bandwidth you need in the worst case (based on the HDs you have or plan to use and the RAID mode you wish to implement) and draw your conclusions.
Read carefully the documentation about the SCSI protocol (www.adaptec.com is a good source to start with).

This is not a matter of raw frame rates in Quake3.
Use your brain, find your needs and plan accordingly.
The satisfaction you can get is far bigger than fragging your opponents in 3D games (I love them!!!!).

I would be glad to help you, if you need.
Bye now,
EnTiTy

[[hd580]]

Ok, I'm no SCSI genius, so let me ask this: What if I had a RAID card w/ multiple channels? Would hooking up Hard Drives to each channel be equivalent to hooking them up to their own SCSI card?

[entity]

Correct, in theory.
But be aware that many low-end SCSI controllers claiming to feature a two-channel architecture (i.e. the Adaptec 2940U2W) actually are a "fake", meaning that you can connect a wider range of peripherals (Fast, Wide, LVD) but NOT to both the channels.

Infact doing so, you force the controller to work at the speed of the slowest protocol.

For example, in the 2940U2W case, you have:

- 1 LVD connector (up to 80MB/s);
- 1 UltraWide connector (up to 40MB/s);
- 1 Fast/Ultra connector (up to 40MB/s);

Do not even think you have a maximum bandwidth of 80+40+40=160 MB/s !!
Real world test I personally made, demonstrated that an UltraSCSI CDROM (Ultraplex 40MAX) connected to the Fast/Ultra port makes the LVD connection drop in terms of max. bandwidth.

Furthermore, if you can, avoid LVD connections unless you like to run into big hassles for trying to terminate the SCSI-chain.
It's often a pain in the ... , and LVD active terminators/cables are ****ing expensive !! And you'll need them all!

My suggestion is to buy a real multichannel adapter or two UltraWide controllers ("wasting" in this case a PCI slot); infact they give you 40 MB/s per controller and currently no HD is even approaching this transfer rate (I have 2 Adaptec 2940UW).
For RAID configuration, things are getting a bit more complicated, due to the high bandwidth involved.

Anyway remember: planning is the key! THINK well first and then go buy!

Glad to hear from you.
Bye now,
EnTiTy

[Stan]

On a multiple channel RAID controller, if you setup one HD per channel, you are still using the same controller.
So, it is not the same as one HD per SCSI controller.

You could compare it to the IDE controller. It has 2 channels and you can hook 2 drives per channel (master/slave).
But, all drives/devices are using the same and only one IDE controller...

Stan

[This message has been edited by Stan (edited 07-24-2000).]

[hoods2]

Here we go, I got two LSI Logic SYM22801 dual channel scsi cards and 4 IBM 2.5 7200RPM hard drives and will see what happens. I will not use 4 pci slot for 4 scsi controllers.
I got the controllers at compgeek.com for $104 w/shipping. Never used LSI Logic cards,hope there good.
Thanks entity I may need your help in the near future.

[entity]

You would not need four PCI slot!
If you plan to use 2 HD in RAID mode you could hook them to one controller and the rest to the other.
It depends on the bandwidth you are going to need.

What kind of controller are the SYM22801 ?
I hope they are at least UWide!

Glad to help you.
Bye now and good work!
EnTiTy

[mig]

Well, gents, how about some numbers? Here's my little baby -- I call her "Cheap Thrills" -- she's an EPoX BP6-KS ($129 Dual Slot 1, BX chipset) with only a single 500e in her at the moment. She's running 2k Server, in standalone mode, ADirectory is not installed. She's also not application configured as a "server" -- at the moment, she's carrying the brunt of my workload until I get the Tiger100 I'm bringing online to 100% working capacity, etc -- but her eventual purpose is to be an AD-BDC, File Server (you'll see in a bit), Telephony Server, WINS Primary, running strictly within a fake-IP number environment (don't use the mission critical machine as your internet presence).

Anyhow, blah blah, this box was made to go FAST, and do it CHEAP. My plan is to put little dink 1 & 2G boot drives on all the other machines in my little nest to provide an OS and a swap file and everything else is going to be running over 100BT to the server (source code compiles, program files folder, the works) -- the tests I've seen have been promising, tho there's some wacky info in my benchmarks for which I can hypothesize the basis of the anomaly, but don't really have the means or time to prove it.

Anyhow, in the upcoming benchmarks, ignore the CPU numbers. CPU numbers on a "configured for workstation use" (ICQ, MSDN, OnSale Java**** console) yadda yadda are pretty darn useless.

Pay close attention to the "I:\" and the "O:\" drives in the HD performance listing -- especially to the write speed and to the CPU overhead. I:\ is the IDE RAID and O:\ is the "\\localhost\" mount of a folder in I:\

The benchmarker I use is CLI MKiii SMP, just because it gives me a nice balanced overview taking multiple cpus into acct.

Anyhow, the RAID setup uses 4 drives of sizes 13, 13, 20, and 27. That gives me a RAID 5 volume of 40GB, and simple volumes in the neighborhood of 6GB and 12GB. All 4 drives (2-WD 2-IBM) are 7200 RPM, U/66, with 2MB streaming buffers.

They interface the motherboard via a pair of Promise Ultra/66 cards. The pair of promise cards is interesting because the bios on the cards is flashable (yawn) BUT one card will detect the presence of the other card and use its BIOS to provide for both -- this helps prevent any "weirdness" if the BIOS's aren't identical (you see... if you have to flash BOTH bioses to upgrade, you must open the case, remove one, flash, replace and remove the other, flash, replace, and close -- ICK!@)

Each drive is jumpered as "Single" and each gets their own channel by themselves on the 4 U/66 heads provided by the two promise cards. (In this way, the striping format of the RAID 5 volume minimized the amount of time it spends waiting for a drive to return data -- remember, on IDE, if you're accessing the slave you can't talk to the master, and vice versa).

Last thing to mention is this little thing has only 196 MB of PC100 RAM -- 64 MB short of the recommended memory available for optimal price performance.

So, taking all of that into account, that means these benchmark values are pretty much my day to day just ****ing around values... not some over-tuned OS made for the sole purpose of running benchmarks:

I believe the comparison machine is the P3/600 stock comparitor data included with CLI. SO, taking into account the speed differential and the fact again that this is a daily use covered in slow-down-systemtray-stuff, those read numbers are insane.

Oh, wait, did you actually want to LOOK at the spec output?

Heheh

this other relative
computer computer performance

Dhrystone 2.1 795 1098 MIPS 72 percent
Whetstone 279 439 MFLOPS 63 percent
Eight queens problem 1680 2352 pps 71 percent
Matrix operations 29798 40899 k ops 72 percent
Number crunch 37600 60645 k ops 62 percent
Floating point 3746 6035 k ops 62 percent
Memory throughput 81024 135040 kB/sec 60 percent

Total CPU 67 percent
Total FPU 62 percent
Average 66 percent

Application Performance 67 percent

Drive C Drive C
Read max 10240 8000 kB/sec
Read average 9126 7584 kB/sec
Read min 4616 6305 kB/sec
Write max 8858 4890 kB/sec
Write average 7062 4769 kB/sec
Write min 5553 4680 kB/sec
CPU usage 0 0 percent

Drive D
Read max 9642 kB/sec
Read average 8467 kB/sec
Read min 4200 kB/sec
Write max 8634 kB/sec
Write average 7388 kB/sec
Write min 6297 kB/sec
CPU usage 0 percent

Drive E
Read max 8619 kB/sec
Read average 6861 kB/sec
Read min 5120 kB/sec
Write max 16357 kB/sec
Write average 11808 kB/sec
Write min 7804 kB/sec
CPU usage 1 percent

Drive F
Read max 8407 kB/sec
Read average 7515 kB/sec
Read min 5960 kB/sec
Write max 7441 kB/sec
Write average 5336 kB/sec
Write min 3680 kB/sec
CPU usage 0 percent

Drive I
Read max 27379 kB/sec
Read average 18156 kB/sec
Read min 8407 kB/sec
Write max 7987 kB/sec
Write average 6259 kB/sec
Write min 4096 kB/sec
CPU usage 12 percent

Drive O
Read max 36571 kB/sec
Read average 33268 kB/sec
Read min 29767 kB/sec
Write max 1110 kB/sec
Write average 977 kB/sec
Write min 897 kB/sec
CPU usage 12 percent

Drive Y
Read max 9110 kB/sec
Read average 8150 kB/sec
Read min 6424 kB/sec
Write max 16410 kB/sec
Write average 13295 kB/sec
Write min 9642 kB/sec
CPU usage 2 percent

Drive Z
Read max 13128 kB/sec
Read average 8739 kB/sec
Read min 5372 kB/sec
Write max 19320 kB/sec
Write average 15489 kB/sec
Write min 10240 kB/sec
CPU usage 1 percent


Just remember,