PDA

View Full Version : Raid... onboard= bad???


blitz
17th February 2002, 10:05 PM
Well, I was talking to a few people, and they said onboard raid is horrible, and you really should get a controller to do a good setup. Is that true with raid0?

Mad Bad John
17th February 2002, 10:47 PM
I too would like to hear peoples opinions on this, as I am planning on a raid 0 setup on my next system.

Imput anybody:xmad2:

Regards MBJ

sptw
17th February 2002, 11:04 PM
That's not true.Most of the mobo manufacturers today use to use a Highpointer or Promise chipset on your mobo.Both two are really good, and doing what they promise.The only negative point is that it has a lot of CPU usage, but who cares?Athlon or Pentium processor today have a excellent power of processing, after all you pay for it, then i think it's better to put it to do a heavy job :D,and don't forget, think about the cost:).For use with Raid0, Raid 0+1 and Raid1 i can't see no problem using on board solution.It has the same performance compared to Raid Card.Now, if you are looking for a heavy and serious job as a server, internet server, backup server or others implementations that need reliable, fault tolerance, hot swap, high level of security as raid5, choose a serious raid card and SCSI implementation.

Mad Bad John
17th February 2002, 11:12 PM
Thanks to "Our man In Brazil":D
Regards MBJ

sptw
17th February 2002, 11:23 PM
Thank you Mad for you Kindness :)

blitz
17th February 2002, 11:38 PM
ok, so the major problem is just that it uses too much cpu. Ok, that makes more sense.

blitz
19th February 2002, 06:12 PM
Well I talked to a lot of people who know a lot on the subject.. and the say it's probably the best idea to get a 3ware controller (the ata33/66 is fine, no need to spend the extra $100 or so for ata100, since the HD won't transfer that fast anyway)

jozzer
19th February 2002, 07:10 PM
Well....on a standard ide controller you see a fair difference between ata 66 and 100.......You sure you don't mean ata 100 and ata133 blitz?

NorthernYankee
19th February 2002, 08:04 PM
Originally posted by blitz
Well I talked to a lot of people who know a lot on the subject.. and the say it's probably the best idea to get a 3ware controller (the ata33/66 is fine, no need to spend the extra $100 or so for ata100, since the HD won't transfer that fast anyway)

Well could you please let me know who you have been getting the info from so that I can inform them correctly so that they do not keep giving people incorrect info:D

A RAID 0 setup with high-speed HD's can transfer data in speeds in excess of 75Mb/s so ATA-100 is beneficial. Now as for ATA-133 RAID there is no real benefit from it except for the insanely large HD sizes this standard allows. And as was said by SPTW The onboard RAID on most Mobo's are either Highpoint or Promise and these are very High quality chipsets, that will allow RAID 0, 1, or 0+1 which is all most home users will need. Now if you need it for a Server that will need Parity checks or such that RAID 5 and others have then I would recommend an Add-on card. As for the CPU usage we are not talking large amounts we are talking in the range of 3-4% CPU usage which is not much when you are using any processor over say 1gig.

But if you wanna spend an extra $60-150 on an add-on card more power to ya but in my eyes you will not get the performance that that money could buy you with other upgrades.

--NY

lechumbl
19th February 2002, 08:09 PM
Hi NY,

Well put!

Facts always speak louder than anything else.

Thanks to both you and SPTW for the good info.

Take care..........

Player0
20th February 2002, 05:05 PM
Backing NY and SPTW here.

1.) There is really no difference between an Add-on PCI Raid controller and the onboard ones. Take the Promise FastTrack ATA100 TX2 controller. Its the same chip as on the Asus motherboards with the onboard raid (or Iwill i think). It simply is on a PCI card and not soldered to the motherboard. Since its the same controller chip, on the same bus, the cpu utilization will be the same!

What your friends are mistaking, are 'hardware' raid controller cards. These cards are much more expensive (above $300 usually), and have their own dedicated 'cpu' of sorts which takes load off the CPU. These cards usually have accellerated XOR capability, which allows them to use RAID5. These cards *are* better than the onboard raid on say the KR7A.

However, the cheap $50-$100 RAID PCI cards, like I said, are exactly the same thing as onboard raid. So if you can get a Promise or a HPT37x chip on your motherboard, go for it. Save a PCI slot and some money.

2.) When using Raid0, the difference between ATA66 and ATA100 is big business. Any single hard drive may beable to burst upwards of 35-40MBs. Thats under the ATA66 spec, so having ATA100 doesnt matter. But when you combine two of these drives, well you can burst upwards of 80MBs. This requires ATA100. If you only have an ATA66 raid controller, your bursts are going to be limited from 80mbs to under 66mbs.

When seriously overclocking my FSB, i will often turn down my ATA100 hard drives to ATA66 or ATA33 (for testing purposes). This improves stability with RAID0 with the higher FSB, but you really do notice the lack of performance.

I hope this helps :)

mdzcpa
21st February 2002, 01:01 PM
I'm late to the party here, but I think sptw, NY, and Player0 are all dead on the money. For a good bang-for-the-buck RAID solution, the on-board controllers are great. For the typical enthusiast Player0 summed it up best:

"So if you can get a Promise or a HPT37x chip on your motherboard, go for it. Save a PCI slot and some money."

That's MOHO too.

Mr. Thompson
28th February 2002, 05:49 PM
You might want to consider one of the new fast IDE drives with an 8 Meg cache. IMHO, RAID 0 is an over used fad. It looks great in benchmarks but unless you work with large files all the time, a single fast drive will provide better performance in daily use.

RAID 0 is not what I would call a reliable storage system for valuable data. Even RAID 0-1 is not secure. While I give the IBM 75GXP Deathstar credit for this, I lost two drives in my RAID 0-1 array at the same time. This took out the whole array.

The best reason to use the on board RAID controlers is in a RAID 1 configuration. While you will loose some performance, it's the only true RAID configuration offered on a basic contoler. Unless you use Deathstars, RAID 1 should provide a little piece of mind if you don't do regular backups.

If you really want to have screaming disk performance, pick up a 15000 RPM SCSI drive and a controler card. Of course this is expensive but the extra reliability of SCSI drives is a comfort.

lechumbl
28th February 2002, 11:53 PM
Hi Mr. Thompson,

Funny you should mention SCSI.
I just got two more 18.2 Gb SCSI H/D's, now making a total of 3 18.2 and my single SCSI of 36 Gb.
I just also scored a Mylex SCSI Controller Card.
Will be making a RAID 5 setup this evening, and making my 36 Gb the image drive.

Will let ya know the results in a few days.

Take care........

Player0
1st March 2002, 04:27 PM
I have to disagree with Mr. Thompson here. Although I do like his choice on the 8mb cache hds :) I currently have dual WD1000BB-SE (thats 100g, 8m cache) drives in Raid-0, and they fly.

I have been using IDE Raid since the spring of 2000, and I have *never* lost data because of Raid0. This is not just my home system, but many business systems I have created, including some file and web servers (which i use Raid 1-0). In fact, one my main file servers, using 4x 15g IBM75GXPs lost a drive, but I was able to recover all data. This one used onboard raid, on a KT7.

"A single fast drive will provide better performance in daily use." That is absurd, im sorry. How can that be possible with over a RAID0 system? I can understand a claim that a raid0 array offers no MORE performance than a single drive, but to claim that raid0 is in anyway slower is just plain wrong imho.

With that said, both my clients and I notice a big performance difference, and I dont mean benchmarks. Boot times, shut down times, all radically improved. Any sort of file or text searching. And of course, large file copying. Games themselves aren't improved much, but game load time is much faster on a RAID machine. Photoshop and other art programs are much improved.

Now, if you have a sub $800 internet machine, which is used for email and occasionaly using Word, than no...raid isnt for you. But if you spend any sort of time doing any sort of real work on a raid machine, you come to appreciate it very quickly.

With that said, SCSI is much better, if you can afford it. I just installed 4 web servers with U160 15k 18.2g hard drives, and those mothers fly. Id love to get a few of those in Raid5 mode. Unfortunately, the price tag is incredible (nearly as much as my 200g raid0 array).

But, not everyones a beleiver ;)

lechumbl
1st March 2002, 04:32 PM
Hi Player0,

Well put, in a polite way.
No way can I agree with Mr. Thompson also.
The purpose of RAID arrays is to insure speed and data integrity.
In fact, I did all of the RAID testing of Digital, now Compaq computers for a year and a half. I speak from much experience there.
Can't wait to get my RAID 5 up and running.
Will let ya know the RAID 5 results soon.

Take care.......

Mr. Thompson
1st March 2002, 05:06 PM
lechumbl - Obviously you have a lot more RAID experience than I. :) My IDE RAID experience is limited to a Promise Fast Track 66 card and the onboard AMI and Highpoint chips. Outside of loading programs and large files, a single fast drive has given me better performance in the office apps I use most. Many of the folks at Storage Review tend to share my opinion. My solution was to go SCSI. I do have one SMP box with SCSI RAID 0. No problems with it at all but then I have never had a SCSI drive fail. By contrast I have lost seven IDE drives, two of them were Deathstars.

Colin

lechumbl
1st March 2002, 05:18 PM
Hi Mr. Thompson,

Please, call me Lee, most all do on NM.

I have found that IBM drives are less reliable than any other I have ever used/tested. Both in RAID and in single use.

They have a higher failure rate than the other brand.
I am sure someone will come back and say " I have IBM, and I have never had a problem or a failure." To that I say, consider yourself lucky. Facts are facts.
I would not own one, and that is from a testing point of view. I have seen way too many bite the big one.

Take care.........

Mr. Thompson
1st March 2002, 05:52 PM
Lee,

Deal. You can call me Colin if you like. It's usually taken as a username on most boards so I use my last name.

Regards,

Colin

Player0
1st March 2002, 08:16 PM
I used nothing but Deskstars for like a year before all this news about failures came though. Maybe out of 30-40 drives, i've seen 3-4 go bad. 1-10 ratio. Not very good. That also means I have 30 potential computers out there that I may get a bad service call on at anytime. I have gone back to Western Digital, since their recent drives have proven reliable, and fairly speedy, if not the top of the pack. StoreageReview seems to like them, so they are my new choice right now, IBMs scare me. Even tho most seem to work very well.

lechumbl
1st March 2002, 08:44 PM
Hi Player0,

Hopefully, since the bad reviews and failures. IBM has made some changes in their new H/D's.
Not much we can do about the existing one, except pray a lot to the HD God.

WD makes good drives and so do Seagate.

All my SCSI drives in my system and the IDE HD's in my wife's system are Seagate, and they seem good for the money.

Take care, and hope ya don't have to replace any drives.........

blitz
2nd March 2002, 01:56 AM
1.) There is really no difference between an Add-on PCI Raid controller and the onboard ones. Take the Promise FastTrack ATA100 TX2 controller. Its the same chip as on the Asus motherboards with the onboard raid (or Iwill i think). It simply is on a PCI card and not soldered to the motherboard. Since its the same controller chip, on the same bus, the cpu utilization will be the same!

The Promise FastTrack ATA100 TX2 isn't 'hardware raid'... it's just as good as an onboard raid controller... I'm pretty sure the supertrak and tx4 ar partially hardware... and the 3ware ones are true hardware raid.

And as I said, the 3ware ata33/66 is first, $100 (maybe a little more), ALL 3ware cards are TRUE hardware raid... and they are master cables only (no slave)... and no single HD today hits, or goes above 66mb/sec, so technically, that card is ideal for now.


Hope that makes sense.. and I'm 95% sure that I am right here.

sptw
2nd March 2002, 09:11 AM
Hi blitz.

Let's go the facts:

1- "The Promise FastTrack ATA100 TX2 isn't 'hardware raid'... " That's True.
2- "it's just as good as an onboard raid controller..." That's true.
3- "I'm pretty sure the supertrak and tx4 are partially hardware..." Supertrack is a hardware raid and TX4 not.To be or not to be, this is the question :).When we said "Hardware card" it means that it has your own processor.It doesn't need to make use of the CPU on your motherboard, besides others implementations.There's no partially hardware in this case.
4- "and the 3ware ones are true hardware raid" Maybe, i don't know all the products from 3Ware.
5- "And as I said, the 3ware ata33/66 is first, $100 (maybe a little more)" I believe in you.
6- "ALL 3ware cards are TRUE hardware raid... and they are master cables only (no slave)..." Look for #4
7- "and no single HD today hits, or goes above 66mb/sec" Well, here we have to make it clear; no single IDE HD (SCSI is another story) today can offer a 66MB/s of SUSTAINED TRANSFER RATE.Don't forget that the interface of HD comunicate with the HD and the computer using a thing named BUFFER.ATA/66 can handle 66MB/s and ATA/100 can handle 100MB/s.
8- " so technically, that card is ideal for now." Sorry to disagree blitz, but it doesn't makes a sense.Actually, now all of us have a very good chance to watch hardware components developing rapidly. Their performance improves and grows incredibly fast, so that we hardly manage to make the parts of our system match each other in this respect. Besides, beautiful games, large audio and video files have turned more predominant recently and are now occupying most of the system storage space on every PC. Indisputably, the connection between the controller and the storage device should keep pace with all these enhancements in order not to stymie the system. The hard disk drives internal data rates ideally shouldn't surpass the burst transfer rates of the interface supported by the system, otherwise the host-to-drive bus will appear the bottleneck, because it will be unable to cope with so much data at a time. As a result, the data will be queuing in the buffer memory of the HDD and slowing down the system performance. One of the possible ways-out in this case, which may seem the easiest, is to increase the size of the HDD buffer memory(ATA/66 and ATA/100), so that the data could be held there before the transfer is successfully finished. However, it will hardly help to eliminate the problem completely, because the higher gets the HDD internal data rate, the more buffer memory is required to avoid delays.However, it is also possible to approach this problem from a different side: to enhance the bus between the harddisk and controller and to increase its bandwidth, so that the data could be transferred at a higher rate. In other words, the data won't be kept in the hard disk drive buffer memory for long and will be transferred via the faster bus almost as soon as it gets there. That's why there will be no need in large buffer memory and it won't require any additional spendings from the HDD developers.Having weighed all pro and contra, Quantum developed this ATA/100 interface with additional speed, which allows transferring the data at 100MB/sec along the host-to-drive bus and hence to unload the HDD buffer memory.Besides that you have to think on this new large LBA(ATA/133 controller that support the new 160GB from Maxtor) , new faster UltraDMA modes with the bandwidth of up to 133MB/sec, HDD noise reduction, a Cyclic Redundancy Check (CRC) feature that's better implemented than on ATA/33 and ATA/66 protocol and others very importants things.Talking about the read speed, in order to give you a better idea of how linear read rate can influence the disk performance, looking at the data transfer rate taken in the beginning and in the end of the HDD with different protocols. The various tests clearly shows that since the liner read rate is higher than the bus bandwidth, when working with ATA/33, the bandwidth appears insufficient.Let's take an example of the IBM 75GXP.This Hd can offer a 444Mbits/s or 55,5 MB/s of data transfer rate.For a ATA/66 interface it is almost on the limit.Besides that, you have probably noticed that there is no difference between the linear read rate in case of ATA/66 and in case of ATA/100. It simply means that ATA/66 is quite enough to provide the appropriate linear read rate, but the difference between them still exists. Namely in real life applications when the HDD is requested more randomly, the data can be read from the disk buffer memory as well as from the disk itself. In case the data is read from the fast HDD buffer, the bus bandwidth turns the limiting factor and the advantages of ATA/100 prove indisputable.Well, if you got really attracted by all these promising details, then you should definitely try it in practice.Don't waste your money on a ATA/33 or ATA/66 controller. However, your system should meet certain requirements, otherwise you will fail to enjoy the thing. Here you are:

You should use an external controller card supporting ATA/100/133 interface or an integrated controller on your mainboard;
Your hard disk drive should support ATA/100/133 interface;
You should have an 80-conductor cable like the one used for ATA/66 interface connection.

For the last, look for this benchmark, and tell me, how can you expect that an ATA/33 or ATA/66 can handle this Higher numbers?
This an IBM 120GXP on Raid0/16K/16K using FastTrack TX2.

blitz
2nd March 2002, 05:24 PM
OK, well I was looking at it from an economical view... as in, if your goign to be buying a $100 raid card, your probably not going to be buying two 160 gig Maxtors. My choice drive is the Maxtor 40gig D740X... that is ata133... but for a 40gig, ata133 is pretty much a marketing scam, right?

And like I said, anyone getting a $100 raid card instead of a $250 is probably not going to the top HDs. Though I do see some of the benefits of the ata100... but for the most part, I don't think it's needed, nor worth that much more money.... on the other hand, I'm guessing the ata100 cards from 3ware also use newer (hopefully better) technology, and might have some performance increase not due to just the ata100 support... and may be worth buying because of that.

lechumbl
2nd March 2002, 05:28 PM
Hi All,

With "normal" RAID needs, on board RAID is enough for most.
If you get the monster large IDE drives, then perhaps you need an add on card.
Remember, mobo's are coming out with ATA133 now, so if you feel the need to spend more money, send it to me on my Pay Pal account.

Take care.........

Player0
2nd March 2002, 06:05 PM
I think that Im not sure what this arguement is about anymore :) I just spent around $450 on my two Western Digital 100gb 8m cache hard drives. And I am using the onboard HPT372 raid on my KR7A-133R. So I dont necessarily agree that I am going the 'budget' route by using onboard raid. I actually see very little real benefit by going to a 'hardware accelerated' raid card. With a AthlonXP running at 2ghz, the CPU utilization of the HighPoint controller is probably going to be a max of 2-3% with my WD drives, and its only that high because I have 16m of read/write cache with those drives, which allows higher bursts.

Although I do agree with most of what Sptw sais in his VERY thourough post, one thing I do know is that not all 'hardware accelerated' raid cards are exactly a like. Even the most robust of these cards still rely on the CPU for some problems. At least for IDE Raid. SCSI is a different story, because there are many more SCSI raid cards, and the interface itself is easier to 'hardware accelerate'. ATA is not as well constructed as SCSI, so its hard to optimize for. With that said, its up to the individual raid card manufacturers to create optimizations in their raid chips. And some of these chips will offload more work from the CPU than other chips. If storagereview still has their database with Raid card CPU util times, you can go see that some cards work better than others. Some cards do a better job of offloading the CPU, others are mixed. So I really dont think its necessarily an all or nothing situation. There are at least a few shades of grey in between.

I am also not aware of 3ware products. Although, I can't find much information either. If they offered lower end RAID cards before, they must be discontinued (which would explain the cheap price). The only product i see listed on their site is a 64-BIT PCI IDE Raid5 accelerator card. Which probably is very fully hardware accelerated, with XOR accelleration for Raid5.

Anyway, I like IDE Raid. I use it for file and db servers for clients, graphics machines, and these days, since its become so inexpensive, just about every machine i build for myself, friends/family, or for clients. I have used some PCI versions of these cards, but have found no benefit to using them over onboard raid if it was available. I infact have an ATA66 Promise raid card sitting unused in a box here, because ATA66 is very limiting these days for RAID0.

The only time I ever recommended a PCI card was the Promise SuperTrak (i beleive), which was for BKPauls file server. I think he wanted like 8 Maxtor 160g hard drives in some sort of RAID config. Im actually not sure now what he ended up using :) The only benefit of this for him was better ATA throughput because the card has more channels, and when using 6-8 drives well, ATA limitations become quite apparant.

Other than that, if you wanted something beyond onboard raid, the only next logical step would probably be SCSI :)

NorthernYankee
4th March 2002, 02:24 PM
Originally posted by blitz
OK, well I was looking at it from an economical view... as in, if your goign to be buying a $100 raid card, your probably not going to be buying two 160 gig Maxtors. My choice drive is the Maxtor 40gig D740X... that is ata133... but for a 40gig, ata133 is pretty much a marketing scam, right?

And like I said, anyone getting a $100 raid card instead of a $250 is probably not going to the top HDs. Though I do see some of the benefits of the ata100... but for the most part, I don't think it's needed, nor worth that much more money.... on the other hand, I'm guessing the ata100 cards from 3ware also use newer (hopefully better) technology, and might have some performance increase not due to just the ata100 support... and may be worth buying because of that.

First to answer your question that for a 40 gig drive is ATA-133 a scam? Well thats a mixed bag here the protocol is for extremely large drives, and 133MB/s transfer speeds. So for the sake of our discussion lets just look at the second point. In most cases ATA-133 vs ATA-100 will not show a noticeable difference. Now I said most in situations of well configured RAID 0 arrays using ATA-133 I have seen burst speeds in excess of 100MB/s but that is the exception not the rule, and with drive technology changing very fast these days I believe you will see 100+ burst speeds from more drives in the near future. As for the Maxtor D740x drives these are great drives and run the same price as most ATA-100 drives.

As for your second point that its not worth the money, in my opionion is not worth the money to buy a Hardware controlled ATA-33/66 RAID card, As has been said many times over for most day-to-day apps an expensive Hardware RAID is not needed.

So don't cripple a great drive like the Maxtor by putting it on a 33/66 card. Use the onboard or a regular Highpoint/promise card you will not regret it.

--NY

Player0
4th March 2002, 05:20 PM
My new WD drives, in raid0, have a SUSTAINED transfere rate of 96mb/s on the beggining of the drives. That means I burst well above 100mb/s ;)