Odd RAID-5 throughput test using HD-Tach

JustRight · 11th July 2006, 12:37 PM

Hi Guys,

I have recently installed and setup a new Areca 1220 PCIe-x8 SATA RAID controller. I have a single Raid Set configured with a single Volume. It is made up of 6 400Gb Hitachi 7K400 SATA1 HDD's in RAID-5.

When testing the throughput on the array using HD-Tach I got the following graph. (Note that I have included a previous benchmark for my original Promise SX4100 PCI controller which was running 4 of the Hitachis is RAID-5)

The Stripe size is 64K and the NTFS Cluster size is 4K.

Click to view full-sized image!
Hosted by UGBox Image Store

This is the result from the new Sandra 2007a.

Click to view full-sized image!
Hosted by UGBox Image Store

Anyway, it has been suggested that a RAID Array should never show that gradual throughput decline that this graph shows as it gets toward the end of the volume. It should be more even across the entire volume.

Is this what you would expect?

Can anyone offer any suggestions as to what might be causing this sort of behaviour? Could it be Cluster size?

Any tips would be appreciated.

Cheers.

douver · 11th July 2006, 1:02 PM

that just shows that the controller nor pci-e bus is the limit of your array. unlike your old one that is obviously hitting a limit somewhere, you can see right at the end of the drives they drop off slightly on the old one

thats what it should look like essentially

JustRight · 11th July 2006, 1:08 PM

Quote:

Originally Posted by douver

that just shows that the controller nor pci-e bus is the limit of your array. unlike your old one that is obviously hitting a limit somewhere, you can see right at the end of the drives they drop off slightly on the old one. Thats what it should look like essentially

Yep, that could be correct. The old Promise SX4100 was a PCI card, so the approx 110Mb/s was probably it's ceiling.

Just found this on the Areca FAQ, which basically says that after about 75% of the volume the throughput drops to 50%

In fact my setup yields much better performance than that, having only dropped to 50% at 100% of the volume size.

This means that the RAID Volume is setup such that the entire array looks and behaves like a single disk drive. So the volume displays the same rotational delay characteristics of a single drive! Very interesting.

http://60.248.88.210/faq/index.php?a..._id=004&id=113

Cheers.

douver · 11th July 2006, 1:17 PM

if you try accessing different disks from different parts of the drive (to try and equal out the transfer) you're always going to be limited by the slowest drive your accessing, so it makes sense to just sequentially access them all

JustRight · 11th July 2006, 4:12 PM

Well, I've been googling like crazy looking for other examples of HD-Tach benchmarks on Areca 1220's. I've actually found a few, and every single one of them exhibits the same general curve shape as my benchmark above, even when they are completely different types of HDD's. Some curves are flatter, some are steeper.

In general though the throughput I am getting from my setup far exceeds the other benchmark examples I have found, so I'm pretty pleased about that

silver6 · 11th July 2006, 4:34 PM

I do some benchies for you tonight.

JustRight · 11th July 2006, 5:17 PM

Quote:

Originally Posted by silver6

I do some benchies for you tonight.

Thanks mate, that would be good. Remember to include details of the number of disks, stripe size, cluster size, RAID type, and anything else useful you can think of.

Aetherone · 11th July 2006, 5:46 PM

thats pretty impressive throughput there, nice little bit of info to know.

The graph shape suggests the limitation in the drives themselves and nothing at all on the bus is a limit

what mobo have you got the areca in?

FarQ · 11th July 2006, 6:10 PM

Quote:

Originally Posted by Aetherone

what mobo have you got the areca in?

that would be in his sig...

infiltraitor · 11th July 2006, 7:32 PM

IMNSHO HDtach is a load of crap...

i got that on a degraded raid5 array using 128k stripe on a RocketRaid 1820A in a 133mhz PCI-X slot
3x 300GB seagate 7200.8 sata150 drives (full array would be 4 drives)

i have a 2nd array on that same card, using the exact same drives.. its bursts arent even over 200MB/s.

the average read is higher too.

Quote:

Originally Posted by FarQ

that would be in his sig...

not everyone has sigs enabled

JustRight · 11th July 2006, 9:05 PM

Quote:

Originally Posted by Aetherone

thats pretty impressive throughput there, nice little bit of info to know.

The graph shape suggests the limitation in the drives themselves and nothing at all on the bus is a limit

what mobo have you got the areca in?

Hi mate,

Yeah, the more research I do the more I realise that the graph is actually pretty damn good! Even if HD-Tach is a POS.

The system is an ASUS A8N-SLI Premium. The Areca is in PCIe Slot 1 and I have a 7800GTX in the other slot. It all works very well together

Cheers.

silver6 · 11th July 2006, 9:18 PM

Initial Benchmarks

All 16k Strip Size, Formated to 16K Allocation Unit
Around 50% Full on both Array
Using 1.41 Firmware, 1.20.0.13 Driver

Seagate 250GB SATA1 NCQ x 3, RAID5

Click to view full-sized image!
Hosted by UGBox Image Store

Seagate 160GB SATA2 NCQ x 5, RAID5

Click to view full-sized image!
Hosted by UGBox Image Store

chedder&cheese · 11th July 2006, 10:22 PM

What are you complaining about, that is still hellishly fast in my opinion.

MichaelMR2 · 12th July 2006, 8:55 PM

16 x 400GB 1260..some old shitty firmware (139 iirc)
Edit: and only 300gb left

JustRight · 13th July 2006, 8:39 AM

Quote:

Originally Posted by MichaelMR2

16 x 400GB 1260..some old shitty firmware (139 iirc)
Edit: and only 300gb left

Whoa!

That's impressive.

Having lots of drives connected clearly makes a significant difference. The 1260 is certainly no slouch!

I'm pretty happy with the performance of the 1220 with 6 drives attached. But that thing of yours must be a knockout.

Cheers.

11th July 2006, 12:37 PM	#1
JustRight Member Join Date: Jun 2001 Location: Adelaide, South Australia Posts: 258	Odd RAID-5 throughput test using HD-Tach Hi Guys, I have recently installed and setup a new Areca 1220 PCIe-x8 SATA RAID controller. I have a single Raid Set configured with a single Volume. It is made up of 6 400Gb Hitachi 7K400 SATA1 HDD's in RAID-5. When testing the throughput on the array using HD-Tach I got the following graph. (Note that I have included a previous benchmark for my original Promise SX4100 PCI controller which was running 4 of the Hitachis is RAID-5) The Stripe size is 64K and the NTFS Cluster size is 4K. Click to view full-sized image! Hosted by UGBox Image Store This is the result from the new Sandra 2007a. Click to view full-sized image! Hosted by UGBox Image Store Anyway, it has been suggested that a RAID Array should never show that gradual throughput decline that this graph shows as it gets toward the end of the volume. It should be more even across the entire volume. Is this what you would expect? Can anyone offer any suggestions as to what might be causing this sort of behaviour? Could it be Cluster size? Any tips would be appreciated. Cheers. __________________ 1968 Morris Cooper 'S' - Classic not Plastic!

11th July 2006, 4:12 PM	#5
JustRight Member Join Date: Jun 2001 Location: Adelaide, South Australia Posts: 258	Well, I've been googling like crazy looking for other examples of HD-Tach benchmarks on Areca 1220's. I've actually found a few, and every single one of them exhibits the same general curve shape as my benchmark above, even when they are completely different types of HDD's. Some curves are flatter, some are steeper. In general though the throughput I am getting from my setup far exceeds the other benchmark examples I have found, so I'm pretty pleased about that __________________ 1968 Morris Cooper 'S' - Classic not Plastic!

11th July 2006, 4:34 PM	#6
silver6 Member Join Date: Nov 2002 Location: Sydney Posts: 959	I do some benchies for you tonight. __________________ \|Intel 486DX2 100\|Gigabyte GA-486TA\|Trident 9000 512KB VLB\|4MB SIMM\|100MB HDD\|

11th July 2006, 9:18 PM	#12
silver6 Member Join Date: Nov 2002 Location: Sydney Posts: 959	Initial Benchmarks All 16k Strip Size, Formated to 16K Allocation Unit Around 50% Full on both Array Using 1.41 Firmware, 1.20.0.13 Driver Seagate 250GB SATA1 NCQ x 3, RAID5 Click to view full-sized image! Hosted by UGBox Image Store Seagate 160GB SATA2 NCQ x 5, RAID5 Click to view full-sized image! Hosted by UGBox Image Store __________________ \|Intel 486DX2 100\|Gigabyte GA-486TA\|Trident 9000 512KB VLB\|4MB SIMM\|100MB HDD\|

12th July 2006, 8:55 PM	#14
MichaelMR2 Member Join Date: Apr 2002 Location: Sydney Posts: 1,137	16 x 400GB 1260..some old shitty firmware (139 iirc) Edit: and only 300gb left __________________ Currently watching the last season of SG1

11th July 2006, 1:02 PM	#2
douver Member Join Date: Mar 2002 Location: Brisbane Posts: 444	that just shows that the controller nor pci-e bus is the limit of your array. unlike your old one that is obviously hitting a limit somewhere, you can see right at the end of the drives they drop off slightly on the old one thats what it should look like essentially

11th July 2006, 1:17 PM	#4
douver Member Join Date: Mar 2002 Location: Brisbane Posts: 444	if you try accessing different disks from different parts of the drive (to try and equal out the transfer) you're always going to be limited by the slowest drive your accessing, so it makes sense to just sequentially access them all

11th July 2006, 5:46 PM	#8
Aetherone Member Join Date: Jan 2002 Location: Adelaide, SA Posts: 6,241	thats pretty impressive throughput there, nice little bit of info to know. The graph shape suggests the limitation in the drives themselves and nothing at all on the bus is a limit what mobo have you got the areca in?

11th July 2006, 10:22 PM	#13
chedder&cheese (Banned or Deleted) Join Date: Dec 2005 Location: Brisbane Southside Posts: 235	What are you complaining about, that is still hellishly fast in my opinion.