HighPoint RocketRAID 2320 experience...

[tpg0007]

So I like many others was lured by the relatively low price. My situation requires 2 4-disk raid5 arrays, and the next cheapest card with 8 SATA ports costs $400 on Newegg, so I decided to take a chance on the 2320.

The array initialization time is worse than pure Windows software raid. It's been 14 hours and my 4x1T array is only at 36%. If rebuilds take that long I might as well return it. Running with latest BIOS/driver on XP SP3 Intel E2200 with 4 GB RAM. Nothing else is running on the box, and CPU utilization is like nil.

On the other hand it has variable sector size so even 32bit OS can see raid partitions up to 16T. Management software doesn't support authenticated SMTP.

Just wondering if anyone else's ran into the slow initialization problem.

 

[TheTorch]

The reason for you poor performance is probably due to your motherboard. Most boards don't provide even 4x speeds in the PCIe slots. There are usually less number of lanes on the board than there are slots (even legacy PCI slots take up 1 lane). Generally if you have a graphics card in the 16x slot, and your board as a 4x slot, then the 4x slot only runs at 1x. Try removing the graphics card and putting it in the 16x slot.

I had amazing performance on an A8N32-SLi Deluxe (I put it in the second SLi slot), and am now using it in a Rampage Formula (put it in the second CrossFire slot).

Some one else who had an A8N-SLi Deluxe had horrible performance with it in the SLi slot. It was because that board had a total of 20 lanes and 16 lanes were consumed by the graphics card in the first 16x slot and the 3 PCI slots consumed the 3 lanes, leaving only 1x for the second SLi slot.

My A8N32-SLi Deluxe was the first "True 16x" gamer board that gave 16 lanes to each card in SLi, thus the RAID card ran at full 4x.

I now have the 2320 on XP x64. I initialized 5x1TB in around 2 hours (The trick is to set rebuild priority on Highest - It affects initialization. I did a foreground init.). I did a test rebuild and it rebuilt in 2:30 hours; exactly the same time it took to rebuild my 5x250GB :O (this saved me from 3 drive failures over the past year). The 5x250 gave me around 190MBps sustained read/write. The 5x1TB gives me more than 400MBps read/write. Yes read and write speeds are the same. Actually the write speeds are 30-40 MBps faster than reading @_@. Confirmed it on various benchmarking tools.

CPU usage is 0 or 1% even during benchmark write tests (Core 2 Quad @ 3.52GHz). Which means the onboard XOR processor of the 2320 is truly accelerating the writes.

 

[wuzy]

It's been 14 hours and my 4x1T array is only at 36%. If rebuilds take that long I might as well return it.

Somewhere in the RocketRAID management software you should be able to set values like 'Initialization rate', 'Rebuild rate', 'Patrol rate'...etc.

My LSI's MegaRAID management software has them.

Initialization and rebuild rate has got absolutely nothing to do with lack of bandwidth (by lack of PCIe lanes). No data is transferred between the host and the controller. It's determined by the controller (and the management software controlling it).

 

[TheTorch]

Initialization and rebuild rate has got absolutely nothing to do with lack of bandwidth (by lack of PCIe lanes). No data is transferred between the host and the controller. It's determined by the controller (and the management software controlling it).

The 2320 is a semi-fake RAID card. Its nothing but a Sata controller that pretends to be a SCSI device. Previous RocketRAID cards actually used to show the individual harddisks connected to the card if you didn't install the drivers because the OS's built in drivers supported the Marvell controller they used LOL.

All reads and writes ARE managed by the CPU. They array WILL NOT rebuild or initialize on its own outside of Windows without the drivers controlling it. The only reasons it has blazing performance is it has a hardware XOR engine on board which accelerate the parity calculation and thus doesn't consume any CPU time while writing unlike other fake raid cards.

 

[sub mesa]

Its not fake RAID. Its hardware assisted. The RR2300 is a smart design, it has its own chip and its 100% hardware RAID; but when using RAID5 arrays, all split/combine/XOR calculations are offloaded to the host CPU; to save cost for a more expensive chip on the RAID controller.

But its still hardware RAID. Fake RAID would mean that the host system sees all the physical disks and implements the RAID mechanic in the drivers; the hardware performing no RAID-specific tasks at all except the RAID BIOS setup utility. In the case of the RR2300, the host system can never see the physical disks. That logic is implemented in the hardware.

So i would designate this kind of controller as "Hardware Assisted" and if you use it in RAID0/1/0+1/JBOD it will be 100% hardware RAID.

 

[TheTorch]

I tried boot to MS-DOS using a floppy, I can see all the RAID partitions including my RAID5 partitions. Which means the card works even without the drivers. But it was EXTREMELY slow when reading/writing from the RAID5 under MS-DOS.

The 2310 or something is fake raid which just pretends to be SCSI so that the computer cannot see the individual drives, and it has no XOR acceleration either. It fully uses the CPU.

The only thing that I found annoying is that when you create a JBOD with a single disk, its puts its RAID headers on the disk meaning that the drive appears corrupted if you plug it in directly to the motherboard. But if you connect an already formatted disk, then it automatically just makes it 'passthrough' and you can access the data without making it a JBOD. I can't figure out how to make a 'blank' harddisk function in 'passthrough'.

 

[wuzy]

...
The array initialization time is worse than pure Windows software raid. It's been 14 hours and my 4x1T array is only at 36%. If rebuilds take that long I might as well return it. Running with latest BIOS/driver on XP SP3 Intel E2200 with 4 GB RAM. Nothing else is running on the box, and CPU utilization is like nil.

....

Just wondering if anyone else's ran into the slow initialization problem.

The reason for you poor performance is probably due to your motherboard. Most boards don't provide even 4x speeds in the PCIe slots. There are usually less number of lanes on the board than there are slots (even legacy PCI slots take up 1 lane). Generally if you have a graphics card in the 16x slot, and your board as a 4x slot, then the 4x slot only runs at 1x. Try removing the graphics card and putting it in the 16x slot.

....

Initialization and rebuild rate has got absolutely nothing to do with lack of bandwidth (by lack of PCIe lanes). No data is transferred between the host and the controller. It's determined by the controller (and the management software controlling it).

The 2320 is a semi-fake RAID card. Its nothing but a Sata controller that pretends to be a SCSI device. Previous RocketRAID cards actually used to show the individual harddisks connected to the card if you didn't install the drivers because the OS's built in drivers supported the Marvell controller they used LOL.

All reads and writes ARE managed by the CPU. They array WILL NOT rebuild or initialize on its own outside of Windows without the drivers controlling it. The only reasons it has blazing performance is it has a hardware XOR engine on board which accelerate the parity calculation and thus doesn't consume any CPU time while writing unlike other fake raid cards.

Are you even answering the right questions? i.e. Are we on the same page?
I'm taking about 'initialization and rebuild rate' which was asked by tpg0007 and not about controller throughput which is what you're attempting to discuss and is of non-relevance to this thread.

 

[jrst]

Which initialization mode are you using? (As wuzy suggested, this is available through the management app.)

1. Foreground -- This is likely to be faster; zeros the disks (write preset pattern across disks), but you can't use the array until it's finished.

2. Background -- This is likely to be slower; generates parity (read-modify-write across disks), but you can use the array immediately.

You might also want to see how long a rebuild takes before making a final decision on whether to keep it.


p.s. TheTorch -- It supports booting from a RAID array (i.e., BIOS INT13), which is why you can see it in MS-DOS. However, that isn't used in normal operation because the performance would really suck, which is why you still need/want OS drivers.

 

[peggy]

Hello,
The reason for your poor performance is probably due to the background initialization.
I think you can use the foreground initialization and set write back cache.