Serial ATA RAID 5 solution! hot?

rgreen83

Senior member
Feb 5, 2003
766
0
0
Newegg has the RocketRAID 1540 with 4 IDE>SATA converters, 4 SATA cables, 4 SATA power adaptors for 125.99.

What makes this hot is this BIOS update from highpoint that makes the 1540 a RAID 5 capable card, very useful for workstations and those who might be running a file server.

The only other SATA card that is RAID that Im aware of is from 3wareand costs nearly 500 dollars. The 1540 is 66mhz PCI capable to help eliminate the pci bandwidth limitation.

Hope this is hot enough to keep from getting flamed on my first post.:D
 

TerryMathews

Lifer
Oct 9, 1999
11,464
2
0
Considering that the FastTrak 133 TX4 is in the same price range, and only supports IDE drives this is very hot.

Good job, rgreen83.
 

yodayoda

Platinum Member
Jan 8, 2001
2,958
0
86
wow, considering i just sold my Abit Serierril adapter (SATA->IDE) for $20, if you have 4 SATA hard drives you could lower the effective price $80 selling off the adapters =) that would this $45 for a RAID 5 card.
 

vetteguy

Diamond Member
Sep 12, 2001
3,183
0
0
Nice...too bad I dropped fat cash last summer on a RAID 5 solution for my server, this would have been even better.
 

WebDude

Golden Member
Oct 11, 1999
1,648
0
0
About the bios update it says: "This version introduces two new RAID levels - RAID 1/0, and RAID 5. RAID 0/1 arrays can not be created with this revision, though it will continue to support existing RAID 0/1 arrays."

Anyone explain to me the diff. between RAID 1/0 and RAID 0/1?
 

caleb1019

Junior Member
Oct 10, 2001
8
0
0
Wow, that looks like a good deal, I was about to get a 3ware raid5 card... gonna have to consider this one now. Does anyone know if you can use two of these cards for 8 drives in the same raid5 array?
 

pxc

Platinum Member
May 2, 2002
2,001
0
0
This is an OK deal since there's virtually no other software RAID5 controller out there (although it's easy to do in Linux). But be aware that the host CPU will have to generate parity data for the the array. The HPT374 doesn't do the partity calculations on-chip.
 

ianbergman

Senior member
Oct 17, 2001
761
0
0
My understanding of RAID 5 (which I will grant you is quite limited) indicates that it uses a lot of processing power to calculate parity... if that's true, does this have enough on-board power to handle a full RAID-5 array, or is it going to start sucking up CPU time?

Hot deal, though. I've been looking at going to RAID-5 for a long time; maybe i should just make the jump to SATA while i'm at it.
edit: whoops, question answered already.
 

linuxthinker

Diamond Member
Aug 30, 2002
3,786
0
0
Originally posted by: caleb1019
Wow, that looks like a good deal, I was about to get a 3ware raid5 card... gonna have to consider this one now. Does anyone know if you can use two of these cards for 8 drives in the same raid5 array?

No contest, stick with the 3Ware :)
 

gaidin123

Senior member
May 5, 2000
962
1
0
Another solution is the RocketRaid404 card which has support for 8 IDE devices and with the newer firmware it also supports RAID5. So, if the serial ATA isn't important to you, you can get an 8 device parallel IDE RAID card for under $100 from newegg.

I'm just waiting for huge IDE drives to come out in SATA before buying. ;)

Gaidin
 

vetteguy

Diamond Member
Sep 12, 2001
3,183
0
0
Originally posted by: gaidin123
Another solution is the RocketRaid404 card which has support for 8 IDE devices and with the newer firmware it also supports RAID5. So, if the serial ATA isn't important to you, you can get an 8 device parallel IDE RAID card for under $100 from newegg.

I'm just waiting for huge IDE drives to come out in SATA before buying. ;)

Gaidin
That card is weird...it's only got 4 channels, but it supports 8 drives? Most RAID cards I've seen require 1 drive per channel. I spent nearly $250 last summer on a Promise 6 channel RAID-5 card. Of course, that was an actual RAID controller which did the processing on board. This is almost an ATA adapter, since it will still require the CPU for RAID calculations. Still a decent deal.
 

TerryMathews

Lifer
Oct 9, 1999
11,464
2
0
Originally posted by: gaidin123
Another solution is the RocketRaid404 card which has support for 8 IDE devices and with the newer firmware it also supports RAID5. So, if the serial ATA isn't important to you, you can get an 8 device parallel IDE RAID card for under $100 from newegg.

I'm just waiting for huge IDE drives to come out in SATA before buying. ;)

Gaidin

The card may support 8 drives, but with each channel having both a master and slave drive. However, there will be no performance benifit in RAID 0 vs. a four-drive stripe. RAID 1 and probably RAID 5 will take a performance hit.

Basically, the two cards are identical except that the RocketRaid supports both parallal and serial ATA. Making it hotter IMO.
 

rgreen83

Senior member
Feb 5, 2003
766
0
0
Originally posted by: WebDude
About the bios update it says: "This version introduces two new RAID levels - RAID 1/0, and RAID 5. RAID 0/1 arrays can not be created with this revision, though it will continue to support existing RAID 0/1 arrays."

Anyone explain to me the diff. between RAID 1/0 and RAID 0/1?

Try this


Originally posted by: pxc
This is an OK deal since there's virtually no other software RAID5 controller out there (although it's easy to do in Linux). But be aware that the host CPU will have to generate parity data for the the array. The HPT374 doesn't do the partity calculations on-chip.

Yes the CPU will still have to calculate parity data, but the simpler architecture of serial has been shown to significantly reduce CPU usage when compared to parallel. This site has a great review of this.
 

OSULugan

Senior member
Feb 22, 2003
289
0
76
Originally posted by: WebDude
About the bios update it says: "This version introduces two new RAID levels - RAID 1/0, and RAID 5. RAID 0/1 arrays can not be created with this revision, though it will continue to support existing RAID 0/1 arrays."

Anyone explain to me the diff. between RAID 1/0 and RAID 0/1?

The difference is in how the different arrays are layered... and mainly comes into play with restoration and reliability.



Explanation Here
Another Explanation (This one with color, and less math)
 

pxc

Platinum Member
May 2, 2002
2,001
0
0
Originally posted by: rgreen83

Yes the CPU will still have to calculate parity data, but the simpler architecture of serial has been shown to significantly reduce CPU usage when compared to parallel. This site has a great review of this.

Geez that's a lame "review". If that guy is getting 43% CPU utilization from a ATA/133 card, he did something seriously wrong. A secondary RAID/IDE controller built onto the motherboard or an add-in PCI card both run off the PCI bus. In either case, I've never had HDTach show more than 3-5% CPU usage with any RAID controller, and it's been typically more like 1-2% with secondary IDE controllers (i.e. not built into the southbridge).
 

TerryMathews

Lifer
Oct 9, 1999
11,464
2
0
Originally posted by: pxc
Originally posted by: rgreen83

Yes the CPU will still have to calculate parity data, but the simpler architecture of serial has been shown to significantly reduce CPU usage when compared to parallel. This site has a great review of this.

Geez that's a lame "review". If that guy is getting 43% CPU utilization from a ATA/133 card, he did something seriously wrong. A secondary RAID/IDE controller built onto the motherboard or an add-in PCI card both run off the PCI bus. In either case, I've never had HDTach show more than 3-5% CPU usage with any RAID controller, and it's been typically more like 1-2% with secondary IDE controllers (i.e. not built into the southbridge).

RAID 5 is eleventy billion times more CPU intensive. Basically, a checksum has to be calculated for every read and write.
 

markgm

Diamond Member
Aug 23, 2001
3,291
2
81
I don't know if it's my setup, but I just got the rocketraid 404, and raid 5 is about unusable. It takes hours to move a gig over to the array (5 40 gig 7200 maxtors), though reading the data off takes no time. I plan on using it primarily as a backup drive since it has redundancy. Hopefully a fix for this will come out.
 

pxc

Platinum Member
May 2, 2002
2,001
0
0
Originally posted by: TerryMathews
Originally posted by: pxc

Geez that's a lame "review". If that guy is getting 43% CPU utilization from a ATA/133 card, he did something seriously wrong. A secondary RAID/IDE controller built onto the motherboard or an add-in PCI card both run off the PCI bus. In either case, I've never had HDTach show more than 3-5% CPU usage with any RAID controller, and it's been typically more like 1-2% with secondary IDE controllers (i.e. not built into the southbridge).

RAID 5 is eleventy billion times more CPU intensive. Basically, a checksum has to be calculated for every read and write.

What in the world does your reply have to do with the article @ http://www.cybercpu.net/review/sata/ ? Or maybe you missed my warning about the HPT374 several posts above?

Not only is your reply off-topic, it's also wrong. Many "hardware" RAID 5 controllers use an Intel 960 processor that is basically a 60MIPS-100MIPS CPU (about the same processing power as a 486 DX4). Unless you're having a problem like markgm is experiencing, a fast system can handle parity data generation without too great of a performance hit. Look at Linux if you want to see how software RAID5 can run at a decent speed.
 

pspada

Platinum Member
Dec 23, 2002
2,503
0
0
The only real problem is that no one actually has a SATA drive to use this with. Why bother using converters to make your ATA133 operate on a SATA controller, since you are dropping the drive speed down to the ATA133 anyway, there is no serious benefit.
 

TheToddler

Banned
May 11, 2001
66
0
0
Originally posted by: markgmI don't know if it's my setup, but I just got the rocketraid 404, and raid 5 is about unusable. It takes hours to move a gig over to the array (5 40 gig 7200 maxtors), though reading the data off takes no time. I plan on using it primarily as a backup drive since it has redundancy. Hopefully a fix for this will come out.
Man, there won't be a "fix" for that. Your card is using your CPU to do the RAID 5 calculations, and that carries a ton of overhead. Software cards are great for RAID 0 or RAID 1, but get a hardware card for RAID 5.
 

PuckMan

Member
Feb 22, 2003
56
0
0
Originally posted by: pxc
Originally posted by: rgreen83 Yes the CPU will still have to calculate parity data, but the simpler architecture of serial has been shown to significantly reduce CPU usage when compared to parallel. This site has a great review of this.
Geez that's a lame "review". If that guy is getting 43% CPU utilization from a ATA/133 card, he did something seriously wrong. A secondary RAID/IDE controller built onto the motherboard or an add-in PCI card both run off the PCI bus. In either case, I've never had HDTach show more than 3-5% CPU usage with any RAID controller, and it's been typically more like 1-2% with secondary IDE controllers (i.e. not built into the southbridge).

I agree, the test is lame. I don't buy into the overhead difference at all. He may of used one controller (maybe on the MB?) that had crummy drivers (no DMA or poor DMA) and then a card with good drivers (SATA.) Reguardless, this datapoint is not valid. The deltas we see between the two drives performance are also so small this could be contributed to where the data was written on the drive (don't forget that outer cylendars have as much as a 2:1 performance increase over inner ones.)

He was right one one thing. The IBM drives are really bad. So bad that I have to replace my entire RAID system after less than one year. I used to go under the premise that I should not even bother with raid 1 as Windows is usually corrupts the disk, thus, your copy would be bad as well. It is less likely that a disk will die. Well, it is RAID 1+0 for me now! Time to replace 4 RAID 0 disks with 4 RAID 1+0 disks. I am going to choose Seagate as they have always worked well for me. Anyone know of any great deals on 100+GB / 8MB Cache ATA / SATA drives????

 

PuckMan

Member
Feb 22, 2003
56
0
0
Okay, on further analysis...

I have been researching these for a while now and it looks like the best option is the LSI Logic card. Unfortunately it is not yet available! Some of the benefits:

6 channel (4 would be fine...but hey, why not 6)
Integrated 100MHz RAM Cache (64MB)
Management support (no big deal)
80302 Processor
64-bit, 66MHz PCI 2.2 <= the main reason I am waiting
SATA Hot Swap
They have been doing this for a LONG time!

The down side:
Not available today
No idea of pricing

There are no other 64-bit 66MHz cards out there. Some are 64-bit (3-ware), some are 66MHz (Promise and others). None of the others has (or at least advertises) any Cache on the board. By the way, the 3-wire site has benchmarks for Parallel and Serial ATA. The data is essentially identical, thus, there is really no advantage to SATA over Parallel ATA. I was talking to someone about this earlier today. I am not sure if there will ever be a performance reason to use SATA over Parallel ATA as far as performance...at least not for quite a while. Here is my justification (flame me as you please!)

1 - A single drive today has a miximum internal xfr rate of about 85MB/s. Max sustained is 32-58MB/s. This is the value we really care about.
2 - Most RAID adapters (ATA) force 1 drive / channel. I know...Promise offers 2/channel on some cards...I did say most!
3 - 58MB/s < 100MB/s < 133MB/s < 150MB/s, thus, even ATA-100 is sufficient for todays technology
4 - Technology will get better, thus, in the next few years I can see internal xfr rates up to the 133MB/s range
5 - Thus, SATA will not buy you any performance over the next few years.

Now, all of the above is valid with the exception of two points (which may be significant!)

1 - For the RAID adapters that allow two drives on the same channel, the maximum transfer rate you can get from two drives is 116MB/s, which is less than the 133MB/s transfer rate you get with ATA-133, still no bottleneck.
a. From PuckMan's law of thermo-datanics ;-> there is no way you will even get there because:
i. You only get the 58MB/s on the outer tracks of the drive, the drive will get slower as you do I/O from the inner cylendars
ii. Master / Slave sux, thus, you won't get to use all the bandwidth anyhow as only one drive can use the cable at a time!

2 - Arguably, there is a small advantage due to a small cache on each drive. This cache is now typically 8MB. If you have a 4 disk RAID 0, this will give you about 32MB of cache. Of that, depending on the manufacturer, part is used for read and part for write caching. Let's say the manufacturer of the drive uses 1/2 for each (dumb algorithm.) This gives you about 16MB of really fast cache that you can do I/O to. For arguments sake, let's say that we are writing to the drive. The initial I/O of the first 16MB of data could transfer REALLY fast (for arguments sake, let's say full bandwidth.) Thus, with SATA-150 it will take about 107ms to write the first chunk of data. On a ATA-133 bus it will take 120ms, for a savings of 13ms. Now, don't go spending that all in one place! Since human reaction time is about 200ms, it is imposible for us to detect this time delta, thus, we watch cluelessly that our system has slowed by 13ms becuase we purchased ATA-133 instead of SATA-150.

That said...I am probably going to purchase a SATA drive...iff...the price is competitive with ATA-133! Why, because the cabling is so much nicer, there are no jumpers, and that is where the technology is headed. If you are going to SATA to get faster I/O, don't! Some time in the next few years it may make a difference, but not today. My advice is that SATA is a great option if the price is almost the same as you are paying for an ATA-133 drive. If not, wait for it to come down. The technology is better, but the performance is not.

Expecting various flames

The PuckMan
 

Monotaur

Senior member
Nov 5, 2001
388
0
0
Originally posted by: PuckMan
Okay, on further analysis...
....
Now, all of the above is valid with the exception of two points (which may be significant!)

1 - For the RAID adapters that allow two drives on the same channel, the maximum transfer rate you can get from two drives is 116MB/s, which is less than the 133MB/s transfer rate you get with ATA-133, still no bottleneck.
a. From PuckMan's law of thermo-datanics ;-> there is no way you will even get there because:
i. You only get the 58MB/s on the outer tracks of the drive, the drive will get slower as you do I/O from the inner cylendars
ii. Master / Slave sux, thus, you won't get to use all the bandwidth anyhow as only one drive can use the cable at a time!
...
Expecting various flames

The PuckMan

I guess it's a little OT, I'm prety sure that the transfer rate of hard drives in constant regardless of cylindar position.... this is why physical sector size is variable (to allow a specific arc length to contain the same amount of data regardless of position)... but maybe I'm wrong?