Hardware vs. Software RAID-0 Performance

grannyknot

I've been poking around the forums and internet for a few hours, but I haven't really found any solid performance comparison of hardware (I'm thinking 3ware 8000 series) and software (no quasi-hardware intermediaries - pure software) RAID-0 solutions.

This (https://www.3ware.com/kbadmin/attachments/Linux2.6WP_0701.pdf) was helpful, but again there was no comparison to software RAID-0.

I'm rebuilding, a Myth+Asterisk+Storage box, and I would like to find the best solution for turning two 250GB drives into one 500GB drive with good performance. If it helps, the processor is a 2.6 GHz P4 running on an Asus P4P800 Deluxe with 1GB of RAM.

I have seen the RAID Tests page, but it looks at RAID-5, a very different beast.

Any insight would be appreciated. I don't mind spending the ~$200 CAD on a 3ware 8006-2LP, but I would like to know that I'm getting some return on the investment.

Thanks,
Don.

NeddySeagoon · Posted: Fri Apr 22, 2005 11:02 am Post subject:

grannyknot,

There is no simple answer. RAID0 does not impose a huge overhead. Reads and writes are striped and can be overlapped. (Not on primary and secondary IDE devices) With SATA drives my sustained data transfer rate went from 50Mb/sec to 84Mb/sec from one drive to RAID0 pair with kernel.

My SATA controller is on a seperate PCI bus to the PCI slots, so there is no bandwidth contention.
If you use a PCI card, you have at most 133Mb/sec for everything on that bus, that could become your bottle neck rather than CPU load. Check your PCI bus layout with lspci.

Phancy Physicist · Apprentice Joined: 10 Apr 2005 Posts: 193

I just wanted to check in and let you know what i get. I use software RAID0 on my two sata and i get 130MB/s sustainable through-put. However this is assuming i have enough cpu free to do this. i'm running a AMD64 3500+ so i have more than enough juice. Have a huge mount of RAM is key because if your RAM is big enough the kernel doesn't need the swap and this cuts down on drive accesses = less cpu time needed for RAID. I'm uping mine to 2Gb soon. Unless you are doing something that accesses the drive all the time(like a server) then I don't think you need the hardware. Just my experiance and opinion

NeddySeagoon · Posted: Fri Apr 22, 2005 2:50 pm Post subject:

Phancy Physicist,

Hmm .. maybe my 84Mb/sec would go up if I stopped S@H running while I test.
I only have a 32bit +3200 XP Athlon though.

130Mb/sec is almost the PCI theoretical maximum transfer rate. How are your SATA controllers connected?
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

grannyknot

Here's the output from lspci:

0000:00:00.0 Host bridge: Intel Corp. 82865G/PE/P DRAM Controller/Host-Hub Interface (rev 02)
0000:00:01.0 PCI bridge: Intel Corp. 82865G/PE/P PCI to AGP Controller (rev 02)
0000:00:1d.0 USB Controller: Intel Corp. 82801EB/ER (ICH5/ICH5R) USB UHCI Controller #1 (rev 02)
0000:00:1d.1 USB Controller: Intel Corp. 82801EB/ER (ICH5/ICH5R) USB UHCI Controller #2 (rev 02)
0000:00:1d.2 USB Controller: Intel Corp. 82801EB/ER (ICH5/ICH5R) USB UHCI #3 (rev 02)
0000:00:1d.3 USB Controller: Intel Corp. 82801EB/ER (ICH5/ICH5R) USB UHCI Controller #4 (rev 02)
0000:00:1d.7 USB Controller: Intel Corp. 82801EB/ER (ICH5/ICH5R) USB2 EHCI Controller (rev 02)
0000:00:1e.0 PCI bridge: Intel Corp. 82801 PCI Bridge (rev c2)
0000:00:1f.0 ISA bridge: Intel Corp. 82801EB/ER (ICH5/ICH5R) LPC Interface Bridge (rev 02)
0000:00:1f.1 IDE interface: Intel Corp. 82801EB/ER (ICH5/ICH5R) IDE Controller (rev 02)
0000:00:1f.2 IDE interface: Intel Corp. 82801EB (ICH5) SATA Controller (rev 02)
0000:00:1f.3 SMBus: Intel Corp. 82801EB/ER (ICH5/ICH5R) SMBus Controller (rev 02)
0000:00:1f.5 Multimedia audio controller: Intel Corp. 82801EB/ER (ICH5/ICH5R) AC'97 Audio Controller (rev 02)
0000:01:00.0 VGA compatible controller: ATI Technologies Inc Radeon R300 NE [Radeon 9500 Pro]
0000:01:00.1 Display controller: ATI Technologies Inc Radeon R300 [Radeon 9500 Pro] (Secondary)
0000:02:03.0 FireWire (IEEE 1394): VIA Technologies, Inc. IEEE 1394 Host Controller (rev 80)
0000:02:04.0 RAID bus controller: VIA Technologies, Inc. VT6410 ATA133 RAID controller (rev 06)
0000:02:05.0 Ethernet controller: 3Com Corporation 3c940 10/100/1000Base-T [Marvell] (rev 12)
0000:02:0a.0 Multimedia video controller: Conexant Winfast TV2000 XP (rev 05)

NeddySeagoon · Posted: Fri Apr 22, 2005 6:31 pm Post subject:

grannyknot,

All these devices are on the same PCI bus

colo · Posted: Sun Apr 24, 2005 7:41 pm Post subject:

I did some benchmarks on my drives (Samsung SP1614C, 160GB 8MB Cache SATA non-native/bridged, connected via Intel ICH-5)

Phancy Physicist · Apprentice Joined: 10 Apr 2005 Posts: 193

NeddySeagoon,

i have SATA on the north and south bridge of my motherboard. I have them on the south because it is faster.

NeddySeagoon · Posted: Mon Apr 25, 2005 4:53 pm Post subject:

Phancy Physicist,

Would you care to post your lspci ?
I'm just curious. You say you are running 64bit AMD CPU, in which case you don't have a North bridge in the conventional sense because its built into the CPU.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

moocha · Watchman Joined: 21 Oct 2003 Posts: 5722

This may be slightly off-topic, but don't forget the availability criterium when evaluating a RAID 0 solution. If your hardware controller goes byebye, so goes your data, unless you can get hold of another card (and 2-3 years down the line that may become a problem if the manufacturer no longer exists, no longer produces that particular model, or provides no upgrade path). Also, you can forget moving the drives into another system that doesn't have the same controller or can't use it for some reason.
By comparison, software RAID will always be around (you can even use the array on opposite-endian machines ;-)

). Comes at a cost too, of course (CPU cycles...), but for valuable data the lock-in problem shouldn't be easily discounted.
_________________
Military Commissions Act of 2006: http://tinyurl.com/jrcto

"Those who would give up essential liberty to purchase a little temporary safety deserve neither liberty nor safety."
-- attributed to Benjamin Franklin

drescherjm · Posted: Tue Apr 26, 2005 4:24 am Post subject:

There should be very little performance difference between the two setups because raid-0 does not have very much overhead. In my opinion software should have a slight performance advantage and will definitly be more safe to transport between systems. As the above poster said. One of the big problems with hardware setups is that if the controller goes bad you usually need the same type to be able to read your data. It is highly unlikely that a different manufacturer's controller would read the data at all.
_________________
John

My gentoo overlay
Instructons for overlay

Phancy Physicist · Apprentice Joined: 10 Apr 2005 Posts: 193

NeddySeagoon · Posted: Fri Apr 29, 2005 2:36 pm Post subject:

Phancy Physicist,

Thats some seriouly impressive numbers.
Does your box usr 64bit and/or 66MHz PCI ?
134Mb/sec is in exxcess of the theoretical max for a 32bit 33MHz PCI bus
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

drescherjm · Posted: Sat Apr 30, 2005 4:27 pm Post subject:

The controller is built into the chipset which is connected to a much faster bus than regular 32 bit 33 MHz PCI. This is one reason why I said that software raid-0 can be faster.
_________________
John

My gentoo overlay
Instructons for overlay

bendy · Posted: Sat Apr 30, 2005 8:15 pm Post subject:

drescherjm is quite right. On modern via chipsets, the sata raid controller is in the southbridge, which is in turn connected to the northbridge by a "v-link" connection. This is way faster than pci.

This is how you can get 1600+ MB/sec cached transfers.