Building an IDE 100 raid array, need some advice

RamaV

Distinguished
Dec 31, 2007
84
0
18,630
Hola,
Ive finally made up my dam mind about new HDs, the old quesiton as to whether go SCSI or not i finally made up my mind. Im going to go with a pair of the new IBM 75 GXPs, prob 2 45 gig ones. I decided that after hearing all about how well IDE raid arrays are working that it will be good enough to avoid having to go SCSI. Here is where i need help

1) Which IDE raid array cards do you all recomend..?
2) Next, my IDE controllers on my old 440BX are slow, i think only ATA33, so ill need a ata100 controller card, right? Im trying to figure out how the raid array card works, so i dont know if ill also need the controller card as well. If so, then any recomendations as to which product.

All help greatly appreciated

Rama

" He who (BLEEPS) nuns, will later join the church " Joe Strummer, The Clash - "Death or Glory"
 
G

Guest

Guest
Recomendations of certian products, I do not have for you.

The RAID controller you purchase will be all you need for the system. You will not need another controller.

However the IDE RAID controllers allow you to connect 4 IDE Hard Drives (in most cases).
So your system will be limited to 4 IDE Devices to the Controller.

you still can use your slow onboard controllers to add CD-ROM drives or burners if necessary.
 
G

Guest

Guest
Go to
www.promise.com
and look at the SuperTrak 100 IDE RAID Controller. It is somewhat expensive, but it is a true RAID Controller with memory on-board. This board has most of the same features of SCSI boards. It supports all levels of RAID at the IDE level. It supports Hot-Swappable setups, so if a drive fails you don't need to power off the system to replace a drive. I found it on www.mwave.com under controllers for $399. This is true RAID, not software RAID like that of the ABIT-KT7A-RAID, or the Asus A7V133 motherboards.
 

RamaV

Distinguished
Dec 31, 2007
84
0
18,630
Mucho thanks.

Question, you said it has memory on board....does this mean that there will be no (or at least minimal) use of the CPU for HD use..? I know with DMA the cpu doesnt get used as much for IDE as it used to, but in comparison to SCSI, it still takes quite a bit of CPU usage, which is why once again im revisiting my decision to go IDE versus SCSI. THe other factor would be heat..but the newer SCSI drives out there are pretty cool. In particular after reading Toms review of the new fujitsu, it runs almost as cool as the IBM 75GXPs. Of course it costs a bundle more, but i think these HDs may be the ones i have for the next bunch of years..hell im buying them before i get a new MB and CPU.

Thanks again,
Rama

" He who (BLEEPS) nuns, will later join the church " Joe Strummer, The Clash - "Death or Glory"
 
G

Guest

Guest
Yes, it has 16mb RAM and a RISC processor on the board to reduce all system usage of CPU and RAM. This unit will do most of the work, not the computer. As for the SCSI to IDE drives, I would not buy SCSI unless you can absolutely find a true need for it. The speed difference in SCSI to IDE is minimal, noticeable at time, but still is not worth the cost to the individual user, SCSI was not really intended as a end-user product when RAID came about and has since been trying to capture the individual user market, but even Apple Computer no longer uses SCSI and have now gone to IDE and Firewire and their standards. SCSI will probably always be reserved for high-end purposes. I have two IBM 75GXP 45gb drives, and they are fast, much faster than my two year old Seagate Cheetah 9.1Gb 10,000rpm 2mb cache drive. It sucks how quickly top SCSI drives are eclipsed by IDE and newer SCSI drives each year. Save you money and get the fastest IDE drives out now, and with the money you saved not having bought SCSI, you can put against the newer, faster IDE drives coming next year.
 
G

Guest

Guest
The memory on the conroller is just another level of cache extended from the cache used on the hard drives. It still uses DMA to handle data transfers from that cache to memory. I've always felt even on the controller cards the DMA processors were better at lowering CPU usage compared to the Southbridge. I don't know I haven't used the Supertrak, but I'm sure it'll do much better at lowering CPU usage in terms of DMA and RAID. It's a full size I/O card too just like you'll find with most SCSI RAID controllers. (This is close but still not quite the same as a good SCSI RAID if you are considering an array with more than 3 drives depending on the RAID level.)



***Hey I run Intel... but let's get real***
 

RamaV

Distinguished
Dec 31, 2007
84
0
18,630
Im gettin some great info here, so ill keep askin dumb questions and maybe ill have everything sorted out (G) It does look like IDE 100 Raids are as advantageous as SCSI nowadays. There is a really small speed advantage for SCSI, but with heat and sound concerns, i think i can rule it out finally.

Now, with the reliability of the HDs out nowadays, in particular the 75GXP models (i know they havent been around long yet, but IBM does have a good rep for reliable drives) what are your all thoughts about Raid 0 as opposed to say 5. 1 would be a waste of my space (im looking at 2-45 gig 75GXPs)..and possibly 2 more in the future) I do ALOT of mulittasking, and a good amount of gaming (im contracted to EA right now)
You all think with the reliability of the IDE drives out now, you would feel comfy going Raid 0, and getting all the speed benefits that come with it..?

Rama (he who has to read up on his Raid levels again hehe)

" He who (BLEEPS) nuns, will later join the church " Joe Strummer, The Clash - "Death or Glory"
 
G

Guest

Guest
There is no doubt that the ATA RAID solution is by and far the cheaper route to go and the performance is realy somthing to behold out of such relativly cheap hardware, BUT. Keep in mind a few of the problems with an ATA RAID solution the first being that to get the high speed out of the ATA RAID you can't have a second drive on either of the 2 channels on the RAID card, one channel one drive. The reason is that the 2 drives will fight for control of the ATA channel. The second is that you will be limited to the rotational latancy of the ATA drives and they are just coming out with 10K HDAs in the ATA interface where as the SCSI drives have the U160 interface and a rotational speed of 15K, this means that they have and average access time of around 2-3mS. Command quing is another part of the SCSI standard that ATA is still missing and helps to make data fetching allot more efficent. Finaly, look at the total sustained I/O's per second, again here is where the SCSI kills the ATA HDD's. The Cheetah X15 drives can muster almost 300 per second, that is just hauling ASS.

Only you know what your true needs are and if you can use and justify the added performance of a SCSI based RAID system and the costs that come with it but those are some of the points to keep in mind.

I'm building a SCSI RAID5+1 array using 2.5G used Seagate Barracuda drives that I bought for only $10.00 each. 12 drives for only $120.00, that's 25GB. It will saturate both of the UW SCSI channels that the drives are on and that is about 76MB/sec sustained. Not bad for 5 year old drives. and SCSI cards.

Good luck.

Mike
 

RamaV

Distinguished
Dec 31, 2007
84
0
18,630
Thanks Mike, great post...these are exactly the issues i have been struggling with. I had been fairly close to deciding on going SCSI when 2 things happened: 1st, i got the specs and got reviews/benchmark results for the IBM GXPs, and 2nd) someone turned me on in an earlier post to this Promise SuperTrak raid controller.

If you havent taken a look at it yet, you may want to. So far the one fault i can see, and im nitpicking to death, is that while it is a 6 channel controller, as you said if i put 2 drives on one controller they willl be a tad slower as they vie for the controller. If i want to go Raid 5, ill have to put 2 on one controller and have the other 2 on theyre own.
Also from readin about the SuperTrak it does have tagged command queuing
Whil the IBMs dont have a high rotational speed, only 7500, so the seek time is a bit slower, the latency is VERY good. They have sustained transfer rate of 37 mb/s. What i havent figured out is if that number is multiplied by the number of identical drives in Raid 0, as others are. If this is the case, ill be shooting right around what you are. Being im buying these to use for the next say 5 years, i really dont mind spending the money, but only if it is truly worth it. Yes i know SCSI Raid will be a bit faster, but they dont seem significantly so from what im reading of this controller. Plus im feeling more and more comfortable with the idea of going Raid0 with these drives. They wont heat up and are not likely to fail anytime soon.
Lookin forward to hearing any more objections, thats what i like, cause if i cant answer them then i know ive seen some shortcomings.

Rama

" He who (BLEEPS) nuns, will later join the church " Joe Strummer, The Clash - "Death or Glory"
 
G

Guest

Guest
The command queing that they talk of is on the controler and not the drive it's self so we are talking about allot more communication between the drives and the controler with the ATA RAID. Another point is that if you put 2 drives on one channel then it doesn't slow it down a little , the slow down can be very dramatic. Remember, with RAID0 it will be writing only 4KB-1MB sized blocks to each drive at a time and this means that they will be negotiating for the bus allot with smaller block sizes. So this means that one drive one channel should be the rule if performance is the goal. After reading the limited specs on their webpage I see that it looks like the card has 6 ATA channels to get around the problem of 2 drives on one bus, it looks to be a nice card. I thing I would also look at is that I/Os per second that the drives can do, the card will more than keep up with them so that drives are the limiting factor. Overall what the hell give it a try, my guess is that unless you were running a database you wouldn't notice any real difference between the SCSI and the ATA RAID solution. On my setup I will be using a software RAID compiled into the kernel of my duel Pentium Pro 200 box so horsepower won't be in short supply and the cards are a pair of AHA-2940UWs for the drives and a AHA-2940U for the CDROMs and burner.

Keep us posted as to the performance that you get out of the array and how easy it is to configure.

Good luck.

Mike
 
G

Guest

Guest
Keep in mind too that it's only a 2 drive RAID 0 setup. A setup that wouldn't create a lot of cpu overhead or IO traffic. So in this case a lot of the benefits you will gain will just be the added cache on the controller in some programs (you add more and will benefit more types a programs).

Some other points..
1) i960 used on the Supertrak may be a RISC based processor and the same used on Adaptec's U160 RAID controllers. But it performs no where near like RISC processors Mylex uses on their high-end SCSI RAID controllers.
2) Software RAID will scale with the CPU. The hardware raid will not.

So in my opinion, this really makes these controllers a cheaper alternative to SCSI for small servers wanting a RAID 5 setup or more than 2 disk array. I saw this same thing when someone pointed out a Review <A HREF="http://www.athlonmb.com/article-display.cfm?ArticleID=54" target="_new">http://www.athlonmb.com/article-display.cfm?ArticleID=54</A> with the ATTO SCSI PCI (a normal dual channel U160 with RAID software) compared to Adaptec's 2100s hardware RAID controller.

You can also see this for SuperTrak100 compared to a FAsttrak100 along with the adaptec AAA-UDMA in this review.
<A HREF="http://www.ebabble.net/Reviews/SuperTrak100/supertrak100.html" target="_new">http://www.ebabble.net/Reviews/SuperTrak100/supertrak100.html</A>

You'll notice the 2 drive Raid 0 Fasttrak100 does more than well. (Some pretty odd numbers but I think it shows the point of RAID 0 with software implemintation, if you've got a good CPU hardware is more likely to hurt performance).


***Hey I run Intel... but let's get real***
 
G

Guest

Guest
I know what my next RAID controler is going to be after reading that review and doing the math, the ATTO is the one. Great link and the devices used are just what I was thinking of using with the exception of the ATTO controler.

Mike
 

RamaV

Distinguished
Dec 31, 2007
84
0
18,630
Chord, MUCHO thanks for the links--i was really trying to find some reviews and benchmarks for the SuperTrak, and that review gave me plenty of info.

One thing Mike may be interested in, which i couldnt figure out by looking at the card but suspected, which this article confirmed is: the SuperTrak doesnt make you put 2 drives on a channel. Realizing that they would fight for bus time, Promise put on 6 single channel connectors-only one drive per channel~! Great trick Promise, so i can do Raid 5 with only 4 drives without worrying about degredation due to 2 drives on a channel.

Im going to be reading up more going from the links ive gotten, and will post any other interesting tidbits i find out. Like one thing this review says which im confused about in its benchmarks is that it says at Raid0 with 4 drives, the FastTrak card (without the onboard CPU and RAM) actually scored better than the SuperTrak one. Beats me..especially since it should use more CPU.

" He who (BLEEPS) nuns, will later join the church " Joe Strummer, The Clash - "Death or Glory"
 
G

Guest

Guest
You like the ATTO article you should read their Mylex review. <A HREF="http://www.athlonmb.com/article-display.cfm?ArticleID=58" target="_new">http://www.athlonmb.com/article-display.cfm?ArticleID=58</A>

That pretty much puts to rest for me the issue of hardware RAID on a desktop/workstation.

The Supertrak clearly has more issues than just bus congestion, while the AAA-UDMA looks like its Write speed is hit hard several times with only it's two channels (but moreover looking at the 2 disk RAID 0 HD Tach results for the Adaptec it's got some processor issues with writes).. (other than that it does pretty well overall)

>>Like one thing this review says which im confused about in its benchmarks is that it says at Raid0 with 4 drives, the FastTrak card (without the onboard CPU and RAM) actually scored better than the SuperTrak one. Beats me..especially since it should use more CPU.<<

The difference between the business and high-end with 4 disk RAID 0 is probably a bus issue and a little cpu overhead. All the disks aren't having to be accessed as much or sequentially in the business tests (as soon as that occurs the Promise pulls ahead of both). Where you really see some problems with the Supertrak and Adaptec card is in the IOmeter. That Cel 500 and software RAID is trouncing those hardware RAID controllers. (I suspect though I don't know the Adaptec is using the same i960 processor). The two cards clearly do better than the Fasttrack at handling 0+1, but how much would that all change on a 1GHz Athlon or PIII? Probably a lot.


***Hey I run Intel... but let's get real***