PERC 5/i RAID Card: Tips and Benchmarks - Page 168 - Overclock.net

utnorris · 03-20-09

Ok, so I am banging my head against a wall right now with this controller. I cannot get Windows to start it and if I take my RAID 0 set and plug it into the controller, create the VD but do not initialize it since that will erase the data, I get a boot disk error. I tried just connecting one drive up to the RAID card with a cloned copy of my boot drives and it didn't like that. SO basically I do not know what I am doing wrong. Here is what I have tried so far:

1 - Boot to windows with no drives on the PERC 5i - windows cannot start it.
2 - Boot to windows with 2 drives in a RAID 0 setup - windows still does not recognize it.
3 - Move my OS drives over to the PERC 5i and configure a RAID 0, but I don't initialize due to it stating all data will be lost - try and boot with no drives on the ICH10 and it fails to boot to windows.
4 - Tried flashing to the LSI FW and I get a failure due to not being able to find the controller.
5 - Tried updating with the latest Dell FW - I get an error stating it's not compatible with my equipment.
6 - Tried just a single drive on the PERC 5i and it wants me to import the configuration, but then fails.

What am I doing wrong here guys? I am running Vista x64, but I didn't think that would be an issue. Need some help please.

DuckieHo · 03-20-09

Have you tried the pin mod?

DJZeratul · 03-20-09

Im having trouble figuring out how to do any benchmarks with HD Tune Pro... Unfortunately the device shows up as a Dell PERC 6/i SCSI Device and is 2199GB, even though the array is 8 1TB disks RAID0... I am assuming it has something to do with the fact that MBR doesnt allow disks over 2TB, only GPT does... and maybe the way HD Tune was programmed it wont do disks this large...

I can do a full benchmark run, but I am not sure it really tests the full gamut of what the array could be doing? All of my write/read tests are basically flat lines, they dont taper near the end like i thought theyw ere supposed to

If anyone has any ideas please let me know.

DuckieHo · 03-20-09

Quote:

Originally Posted by DJZeratul

I can do a full benchmark run, but I am not sure it really tests the full gamut of what the array could be doing? All of my write/read tests are basically flat lines, they dont taper near the end like i thought theyw ere supposed to

They should taper a bit... you are seeing the first 25% of the performance chart where it is pretty level.

I don't know how to do benchies on GPT...

DJZeratul · 03-20-09

Quote:

Originally Posted by DuckieHo

They should taper a bit... you are seeing the first 25% of the performance chart where it is pretty level.

I don't know how to do benchies on GPT...

Read speeds on RAID 5:

Its odd, HD Tach shows the full sized array

DJZeratul · 03-20-09

This post will be where i put my HD Tune images.

There are a lot of iterations so to start off I am going to do 2 variations of Read and Write for each stripe size (1024K, 512K, 256K, 128K, 64K, 32K) in RAID 5 with 8 drives. These variations will be:

Read with Adaptive Read Ahead
Write with Write Back
Write with Write Through
Read with No Read Ahead

After I finish the RAID 5 x8 disks stuff, I will be moving along with only the performance benchmarks, aka all caching modes on, Adaptive Read Ahead active and Write Back on. Therefore there will only be a Read and a Write test for each iteration.

To speed things along at this point, I will also be analyzing the fastest stripe size after the RAID 5 x8 disk tests are complete. Whichever one has the fastest read/write performance based on my own observations, I will be using that stripe size only. The sets I plan to do are:

RAID 5 with 4 Disks
RAID 0 with 4 Disks, 8 Disks
RAID 6 with 8 Disks
RAID 50 with 8 Disks
RAID 60 with 8 Disks
RAID 5 (DEGRADED)
RAID 50 (DEGRADED)
RAID 6 (DEGRADED)
RAID 60 (DEGRADED)

IF you want to know what something looks like and I havent covered it in this benchmark summary, or have any requests for a specific test, please do so in this thread or PM me and I will do a test for you specifically. Please note that I probably wont be doing benchmarks for very long after I finish my own, because I want to start getting data onto these drives ASAP. But i also want to bench the hell out of them first

So if you do have a request, make it ASAP so I can fit it in over the next few days.

KEY to my Codes:

{# of drives}x {type of drives}. {RAID Level}, {Write Back(WB) or Write Through(WT)}, {Adaptive Read Ahead(ARA), Always Read Ahead(RA), No Read Ahead(No RA)}, {Cached or Direct I/O} {Stripe Size}

================TEST DATE: March 20, 2009, 2:00 AM PST================

STRIPE SIZE=1024K

STRIPE SIZE=512K

STRIPE SIZE=256K

STRIPE SIZE=128K

===================TEST PAUSE - 4:16 AM (must sleep)===================

RobiNet · 03-20-09

Quote:

Originally Posted by ounderfla69

The port replicator lets 5 drive use one Sata lane to transfer data. I have one for sale with a case but they way it works if you plug the replicator into a sata controller that isnt the Silicon Image 3132 or one or two other the only drive it will see is the first volume on the replicator. I have 3 drives on the port replicator the first 2 500GB was in raid 0 and the other was just a single drive. When plugged into the correct card 2 drives were seen the 931GB raid 0 drive and the single 500GB drive. If I plugged it into another Sata controller it only saw the 931GB drive and that was the only drive seen. Why would you want to use a port replicator on a PERC5/i anyway, it has a limited bandwith for all the drives.

So there is no problem in connecting PM to PERC 5/i ? I have ASUS P5Q-E and there is SIL chipset so I can configure PM.
I want to connect few small SSDs - how big bandwith drop should I expect ? 5% ? 20% ? 50% ? more ? And per drive or on whole RAID ?

DJZeratul · 03-20-09

Quote:

Originally Posted by RobiNet

So there is no problem in connecting PM to PERC 5/i ? I have ASUS P5Q-E and there is SIL chipset so I can configure PM.
I want to connect few small SSDs - how big bandwith drop should I expect ? 5% ? 20% ? 50% ? more ? And per drive or on whole RAID ?

No, he is saying he plugged it into a non SIL-xxxx controller and only the first drive on the port replicator showed up. I think the same thing will most likely happen if you use it on a 5/i. i have plugged a port replicator unit into a Marvell SAS controller and it did not work either.

The port replicator, on a compatible controller, shares the bandwidth of a single port between the drives. So if you have 5 drives, each drive gets 20% of the port's total bandwidth (3.0Gbps Theoretical Max). In a RAID 0, you would gain most of that performance back and it would most likely act as if it was a larger single drive, but you would lose the potential that RAID 0 gives you on a real multi port SATA controller, since each drive in the RAID would then get a total of 3.0 Gbps PER DRIVE, wheras a port replicator you get 3.0 Gbps across the entire RAID. At that point, you are better off just buying a large drive because the benefits of RAID 0 are completely lost.

Port multipliers are mainly made for a user who wants to add more storage drives than they have SATA ports for. Definitely not made for RAID.

Also, as you mentioned before about "Command-based switching" vs "FIS–based switching", if the controller/port multiplier uses Command based switching, your RAID 0 will be as slow as a single disk's 20% bandwidth because the OS can only issue a command to a single drive at a time. Therefore when striping data it would be writing to a drive at a time rather than all drives at once like a true RAID 0 setup.

RobiNet · 03-20-09

Quote:

Originally Posted by DJZeratul

No, he is saying he plugged it into a non SIL-xxxx controller and only the first drive on the port replicator showed up. I think the same thing will most likely happen if you use it on a 5/i. i have plugged a port replicator unit into a Marvell SAS controller and it did not work either.

I think you misunderstand ounderfla69 - he says that his non PM SATA controller see only first VOLUME - 2x 500 GB (931GB) - don't see 3'rd single 500 GB drive ... so if I'll configure PM as single 5-drive volume ...

Quote:

Originally Posted by DJZeratul

The port replicator, on a compatible controller, shares the bandwidth of a single port between the drives. So if you have 5 drives, each drive gets 20% of the port's total bandwidth (3.0Gbps Theoretical Max).

ok ... so 20% of 3Gbps = 75MB/s - it is more/equal what single regular drive can handle

Quote:

Originally Posted by DJZeratul

In a RAID 0, you would gain most of that performance back and it would most likely act as if it was a larger single drive, but you would lose the potential that RAID 0 gives you on a real multi port SATA controller, since each drive in the RAID would then get a total of 3.0 Gbps PER DRIVE, wheras a port replicator you get 3.0 Gbps across the entire RAID. At that point, you are better off just buying a large drive because the benefits of RAID 0 are completely lost.

but if I want to put bunch of SDHC->SATA adapters - I don't need 375MB/s transfer per "drive"

Quote:

Originally Posted by DJZeratul

Port multipliers are mainly made for a user who wants to add more storage drives than they have SATA ports for. Definitely not made for RAID.

Also, as you mentioned before about "Command-based switching" vs "FIS–based switching", if the controller/port multiplier uses Command based switching, your RAID 0 will be as slow as a single disk's 20% bandwidth because the OS can only issue a command to a single drive at a time. Therefore when striping data it would be writing to a drive at a time rather than all drives at once like a true RAID 0 setup.

What OS ? we are talking about DELL PERC 5/i - hardware host - so OS have nothing to do ? OS will send command(s) to PERC and becuase there will be 512 KB cache + Write-Back - it will get 0-time response ...

Or am I totally wrong ?

DuckieHo · 03-20-09

Quote:

Originally Posted by DJZeratul

This post will be where i put my HD Tune images.

There are a lot of iterations so to start off I am going to do 2 variations of Read and Write for each stripe size (1024K, 512K, 256K, 128K, 64K, 32K) in RAID 5 with 8 drives. These variations will be:

Read with Adaptive Read Ahead
Write with Write Back
Write with Write Through
Read with No Read Ahead

After I finish the RAID 5 x8 disks stuff, I will be moving along with only the performance benchmarks, aka all caching modes on, Adaptive Read Ahead active and Write Back on. Therefore there will only be a Read and a Write test for each iteration.

I think something maybe wrong.... I would expect 8 drives in RAID 5 to get closer to 650MB/s and not 440MB/s....

		Overclock.net - Overclocking.net > Components > Hard Drives & Storage > RAID Controllers and Software
PERC 5/i RAID Card: Tips and Benchmarks

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)