Sign in with
Sign up | Sign in
Your question

GA-EP45-UD3P RAID 5 Performance (Gigabyte support seems a bit off)

Last response: in Motherboards
Share
November 3, 2009 5:17:45 PM

I am getting terrible RAID 5 write performance (avg 15 MB/s). I'm wondering if I might have a defective motherboard. I contacted Gigabyte and I think the person I spoke with is on crack. He says the ICH10R controller is software raid. :o  See the message chain below. Does anyone else here use the ICH10R controller in RAID 5 mode and what kind of read/write performance do you get?




Message Chain:
----------------------------------------------------------------
Sent : 10/25/2009 14:03
Question : I'm getting absolutely terrible RAID 5 write performance. It's writing at roughly 15 MB/s. I know there are write performance penalties when using RAID 5, but not like this. Does this sound like a defective controller to you?

Gigabyte: Which OS you are using and have you and enable the write-back cache setting?

Sent : 10/30/2009 23:43
Question : OS: Windows 7 x64
Write-caching defaults to enabled.


Gigabyte: That is all needed to be done, which applications you are testing with under Windows 7 and are those utility optimized under Windows 7?

Sent : 10/31/2009 14:35
Question : Everest
I also did some standard file copies using explorer, xxcopy, robocopy, copy, xcopy to/from a ram drive (eliminating the potential bottlenecks from the other drives, drives sharing the same raid controller, etc) I've read some other reviews of the ICH10R controller and it looks like they are getting the same write performance (ouch). However, they are getting much better read performance with similar configurations. Apparently, you're not convinced that it is a hardware fault (and neither am I), but I wanted to get your opinion. I have both burnintest pro and everest and neither of the utilities are reporting any issues when I run their diagnostics. This is just very disappointing performance and I'm just going to have to suck it up or by a dedicated RAID controller. If you do feel that it's likely a hardware fault, let me know and I'll do what ever troubleshooting is necessary to get a replacement.


Gigabyte : since it's software raid it can only perform up to an certain spec. You can obtain an raid controller card to use hardware raid.

Sent : 11/3/2009 04:44
Question : You lost me or I lost you somewhere in this conversation. It's NOT software RAID. I am using the ICH10R integrated controller on the GA-EP45-UD3P motherboard in RAID 5. The Seagate 7200.11 drives average around 70 MB/s write speed when not used in RAID. In your opinion, should I be getting a faster write perfomance than 15 MB/s when using RAID 5? If your answer is yes, I need an RMA issued. Sorry for the confusion.

Gigabyte : The Intel ICH10R is software raid, the only way to monitor and rebuild the raid is within the OS with the utility running. Hardware raid can perform rebuilding within the bios not only in the OS. You can try running SiSoft Sandra as well and compared your results to others
Certain drives are faster than others so perform on different drives will may not be the same.
a c 177 V Motherboard
November 3, 2009 6:37:47 PM

Quote:
It's NOT software RAID

It IS software RAID - your CPU is doing all the work. That's why real RAID uses an IOP (input/output processor - like the Intel 342, q.v.); and the better ones are two core - one core runs the XOR engine to calculate the parity stripe(s), while the second runs the SCSI/SATA protocol stack! The linux world names it correctly, and more expressively - there, it's called "FakeRAID"... When the CPU has to shoulder the brunt of the work, RAID 0, 1, & 10 are pretty simple - it's simply 'splitting' & 'recombining' data streams - so the overall bandwidth gets better, even at the sacrifice of som processor cycles wasted; when you're doing RAID5, the CPU has to handle the whole load of all the parity calculations and writes, at stream speed - which slows things down inexolerably...

I've never bothered to even test RAID5 on my primary ICH9R; if you want speed, RAID0 some fast drives (I use 4 VRs - roughly 200M/S reads on each pair...), if you want data security as fast as practical, RAID1 some reliable RAIDable drives (can't use WD C_Blacks - I have RE3s) and 'waste' a drive - no other way...
November 3, 2009 7:08:19 PM

I'm pretty sure that software RAID is controlled through the operating system.

According to Intel's ICH10 datasheet, it has two controllers for the six ports.

Quote:
The SATA function in the ICH10 has three modes of operation to support different
operating system conditions. In the case of Native IDE enabled operating systems, the
ICH10 utilizes two controllers to enable all six ports of the bus. The first controller
(Device 31: Function 2) supports ports 0–3 and the second controller (Device 31:
Function 5) supports ports 4 and 5. When using a legacy operating system, only one
controller (Device 31: Function 2) is available that supports ports 0 - 3. In AHCI or
RAID mode, only one controller (Device 31: Function 2) is used enabling all six ports.


I don't doubt that those controller(s) may use the CPU for calculations (potentially the reason for the terrible RAID 5 performance), but I doubt that this is considered software raid. ICH10R is the southbridge chipset and I don't know how to identify the RAID controller other than by what is quoted above as Device 31: Fucntion 2. If it was true software RAID, I wouldn't be able to configure it from the BIOS (which I can) and dual booting into other operating systems such as Linux or using winpe wouldn't recognize the array (which they do).
Related resources
November 3, 2009 7:57:49 PM

Bilbat, I think you are likely right on money about the CPU overhead being the culprit. However, for the sake of argument, calling it "software RAID" is also incorrect.

Quote from Wikipedia:
Quote:
Firmware/driver-based RAID

Operating system-based RAID doesn't always protect the boot process and is generally impractical on desktop versions of Windows (as described above). Hardware RAID controllers are expensive and proprietary. To fill this gap, cheap "RAID controllers" were introduced that do not contain a RAID controller chip, but simply a standard disk controller chip with special firmware and drivers. During early stage bootup the RAID is implemented by the firmware; when a protected-mode operating system kernel such as Linux or a modern version of Microsoft Windows is loaded the drivers take over.

These controllers are described by their manufacturers as RAID controllers, and it is rarely made clear to purchasers that the burden of RAID processing is borne by the host computer's central processing unit, not the RAID controller itself, thus introducing the aforementioned CPU overhead which hardware controllers don't suffer from. Firmware controllers often can only use certain types of hard drives in their RAID arrays (e.g. SATA for Intel Matrix RAID, as there is neither SCSI nor PATA support in modern Intel ICH southbridges; however, motherboard makers implement RAID controllers outside of the southbridge on some motherboards). Before their introduction, a "RAID controller" implied that the controller did the processing, and the new type has become known in technically knowledgeable circles as "fake RAID" even though the RAID itself is implemented correctly. Adaptec calls them "HostRAID".
a c 177 V Motherboard
November 3, 2009 10:05:29 PM

You have an underlying misconception here: firmware is software; it is simply 'burned' or 'flashed' into a piece of 'hardware memory', be it EPROM, EEPROM, NVRAM, FeRAM, MRAM - doesn't matter, it's still software in memory... There are two basic ways to run these programs - either the hardware is 'vectored' into the hard storage itself, to execute, using other volatile storage for working register blocks; or, the boot-strap procedure copies the ROM code into volatile memory, and executes there - once again - doesn't matter. Executing directly from ROM is usually fairly inefficient, as timings for VRAM are much faster - but, for a limited system, like a washing machine controller, or a coffee pot, you might get by with the micro-controller's own register memory for 'working' storage, and not need (the expense of) a memory subsystem at all... We just got a Roomba, and I've already downloaded its 'firmware', and am busy disassembling :pt1cable: 

Just because something is 'in the BIOS' gives it no special powers (or any particular 'boost' in execution speed, except for that it was likely done in assembler, and highly sub-optimized), it's just object code running on the processor. The BIOS makes a BIG difference to everything, though, as in the end, pretty much everything else is eventually vectored into the BIOS functions, one way or another... For example, about five or six months ago, Intel updated the ICH BIOS code, correcting some rarely triggered bugs, and obviously offering some better sub-optimization of the RAID functions, as it (in combination with new OS drivers [and only in combination with new OS drivers!]) increased throughput by about 5%; it took quite some time for manufacturers to 'propagate' these changes through their existing BIOS, and, for a while, there were a number of 'hacked' BIOS floating around for the faster platforms (X38 & 48, X58), incorporating the new RAID BIOS while waiting for the official releases...

However, no matter where the execution code comes from, it's still got to 'time slice' with the rest of the system's processor useage; the only way out is exactly the way that RealRAID cards do it - offload the work to another processor, with its own memory subsytem (and the medium priced RAID cards now accept two gig SODIMMs), or sometimes, as I mentioned, to a dual-core...
November 4, 2009 2:44:16 AM

Great explanation. I'm still thinking that the 15 MB/s write speed is dismal performance. The q9550 processor that I run has plenty of cpu cycles available when I'm doing nothing except the file transfer. After reading this article Southbridge Battle: 780a, ICH10 and SB750, Compared I really am thinking that something is wrong. I also noticed that the 7200.11 drives I'm using have a recommended firmware update. So I'm going to move my data off the array tonight and test it using h2benchw tomorrow and see if my results are even close to what the article is showing. I'll be curious to see if the firmware improves performance or not.
a c 177 V Motherboard
November 4, 2009 3:08:04 PM

Ahhh... I'm going brain-dead:
Quote:
Seagate 7200.11

that's the first thing I should have noticed - you likely have the dreaded "Seagate Stutter"; info here:
http://seagate.custkb.com/seagate/crm/selfservice/searc...
Download their ID tool and follow posted instructions carefully - updating firmware on some drives that don't need the firmware is known to render the drives inoperable...
I got a pair of 1.5s that had sequential serial numbers. 'broken off' the original Seagate foam OEM shipping foam thingie - one had the 'stutter', one didn't - who knows??
November 4, 2009 3:41:51 PM

Two of the three drives I have are sequential serial numbers. However, they are all using the same firmware revision. I've read that a few people have "bricked" their drives after applying the firmware update. Seagate claims to have resolved that issue. Only one way to find out.
April 3, 2010 12:33:41 PM

bilbat said:
Ahhh... I'm going brain-dead:
Quote:
Seagate 7200.11

that's the first thing I should have noticed - you likely have the dreaded "Seagate Stutter"; info here:
http://seagate.custkb.com/seagate/crm/selfservice/searc...
Download their ID tool and follow posted instructions carefully - updating firmware on some drives that don't need the firmware is known to render the drives inoperable...
I got a pair of 1.5s that had sequential serial numbers. 'broken off' the original Seagate foam OEM shipping foam thingie - one had the 'stutter', one didn't - who knows??



Bilbat - you are off-the-charts-brilliant. You natively THINK in the logic of how information is processed on a computer. AMAZING. I realize you're presently addressing "seagate stutter," but aside from that, you still had a general expectation that LOWER performance from RAID-5 was not surprising. So, with all due respect (and that's quite a bit), the question that has NOT been answered is:

What would be the value in a manufacturer, Gigabyte in this case, offering a "feature" such as RAID (5 in this usage case, which is performance biased) if the end result is to reduce performance at the cost of drive-space/money? I think, the OP's point is that one would expect, that irrespective of hardware or software RAID implementation, that there would be an advantage for doing so for each method the manufacturer advertises the ability for. What other reason if not performance improvement would someone have for RAID 0 or 5? I would essentially regard it as a "scam" for a corporation to advertise a feature that in fact, were a detriment.

My expectation would be that, at the expense of processing cycles up until the CPU is fully utilized, that the limit of transfer rate (as access time is clearly not improved by RAID and thus shall be ignored) would eventually be the bus speed, which in this case, SATA-2, should be about 300MB per second. Of course, this may take more drives than there are SATA-slots to reach this threshold, and, obviously, the 375MB per second is theoretical. -- BUT, there should be a clear bottleneck, as opposed to an arbitrary one, let alone the case stated by the OP, of detriment without cause.

I'm actually quite shocked to have found this thread, because ultimately, the OP is doing the exact thing I was contemplating with respect to cost-benefit, hinged on the change in transfer rate for the effort, cost, and loss of space. Hope to hear back from you bilbat...

Regards... and again, you have an impressive degree of knowledge! I signed up exclusively to speak with you.

Truman
a c 177 V Motherboard
April 3, 2010 1:25:15 PM

Jeez - I'm blushing [:proximon:1] !

First, re performance - look here:
http://www.tomshardware.com/forum/272698-30-gigabyte-gu...
at the RAID & 'FakeRAID': Speed vs Data Security section...
Quote:
What would be the value in a manufacturer, Gigabyte in this case, offering a "feature" such as RAID (5 in this usage case, which is performance biased) if the end result is to reduce performance at the cost of drive-space/money?

Well - the whole issue (like almost every 'engineering' issue!) is a trade-off between speed and data security. While FakeRAID 5 is painfully slow, I can see situations where it could be useful - say, for a windoze home server, whose main job is to sit in a closet in the basement and back-up every machine in the house sometime between 11PM and 6AM... Don't really care how long it takes (assuming it does get done before the first user has his coffee [:isamuelson:6] and sits down at a system), but really don't want to lose aunt Ida's last family reunion picture, now that she's been 'in the ground' [:lorbat:5] for two years, and won't be taking any replacement pictures!

And it's really not so much a GIGABYTE issue - they are pretty much just 'announcing' the features available from their component parts - and, pretty much, all southbridges have 'supported' RAID5 for years. Now, they could put in a 'caveat' about it, and say "...RAID5 * but, will slow your throughput down to 25M/S *", but you know how the marketing guys are - everything is a 'saleable' feature, seen through their rose-colored glasses - and they don't really know what RAID is! (For what it's worth, I spoke with a marketing type, and he told me they hate the engineering guys, too - I believe his quote was "too much recipe - never any cake!")

If you want a large, fast, 'safe' array, you've just got to do what the server guys do - RAID6 on a dedicated (but pricey!) card, and four drives 'tossed', for two parity disks, and a couple of 'hot-spares'...

As for the 'bottle-necking', you have to remember that data transfer to nearly every 'large model' device is handled by DMA - the processor never has to 'see' the data in 'byte-wise' fashion; it just 'waves it's magic processor wand' [:jaydeejohn:3], and the big 'block' of data that was here, appears there! When FakeRAIDing 5, however, every single byte must be 'parity calc'd, individually, when writing - so writes become despicably slow!
April 3, 2010 10:33:35 PM

bilbat said:
Jeez - I'm blushing [:proximon:1] !

First, re performance - look here:
http://www.tomshardware.com/forum/272698-30-gigabyte-gu...
at the RAID & 'FakeRAID': Speed vs Data Security section...
Quote:
What would be the value in a manufacturer, Gigabyte in this case, offering a "feature" such as RAID (5 in this usage case, which is performance biased) if the end result is to reduce performance at the cost of drive-space/money?

Well - the whole issue (like almost every 'engineering' issue!) is a trade-off between speed and data security. While FakeRAID 5 is painfully slow, I can see situations where it could be useful - say, for a windoze home server, whose main job is to sit in a closet in the basement and back-up every machine in the house sometime between 11PM and 6AM... Don't really care how long it takes (assuming it does get done before the first user has his coffee [:isamuelson:6] and sits down at a system), but really don't want to lose aunt Ida's last family reunion picture, now that she's been 'in the ground' [:lorbat:5] for two years, and won't be taking any replacement pictures!

And it's really not so much a GIGABYTE issue - they are pretty much just 'announcing' the features available from their component parts - and, pretty much, all southbridges have 'supported' RAID5 for years. Now, they could put in a 'caveat' about it, and say "...RAID5 * but, will slow your throughput down to 25M/S *", but you know how the marketing guys are - everything is a 'saleable' feature, seen through their rose-colored glasses - and they don't really know what RAID is! (For what it's worth, I spoke with a marketing type, and he told me they hate the engineering guys, too - I believe his quote was "too much recipe - never any cake!")

If you want a large, fast, 'safe' array, you've just got to do what the server guys do - RAID6 on a dedicated (but pricey!) card, and four drives 'tossed', for two parity disks, and a couple of 'hot-spares'...

As for the 'bottle-necking', you have to remember that data transfer to nearly every 'large model' device is handled by DMA - the processor never has to 'see' the data in 'byte-wise' fashion; it just 'waves it's magic processor wand' [:jaydeejohn:3], and the big 'block' of data that was here, appears there! When FakeRAIDing 5, however, every single byte must be 'parity calc'd, individually, when writing - so writes become despicably slow!



Thank you for the fast reply! :)  And I just want to reiterate your position (and likely reality of the matter) that even RAID-0, which is exclusively for speed, if implemented via OS level RAID mgt, will still be slower than a single drive's throughput - essentially making the ONLY purpose for SW based RAID, fault-tolerance via redundancy?

I used to (waaay back when) use a Mylex RAID adapter... and then switched to a an Adaptec - neither lived up to my throughput expectations, and certainly introduced management overhead. My goal is to have a fast OS/program drive (data security is unimportant because I reinstall frequently and keep "special folders" like My ___ pointing at my data drive), which is seemingly going to be better served by SSD than OS based RAID, and even again, SSD would win in an efficiency battle against buying a RAID adapter because, they cost [almost] the price of an SSD! And after buying said adapter AND using stripes as parity, I'm essentially back at the cost of an SSD, only retaining the conventional drive's latency... And ultimately, an SSD gets close to maxing out the BUS speed as it is! Right?

Do I follow you correctly?

I am however using (temporarily) an external RAID box my data... and intend to upgrade to a NAS, probably the Seagate BlackArmor if costco starts carrying it again; being able to return something for literal eternity is too good a protection to pass up...

Please do excuse my wordiness, I find in echoing what I gather someones position is and how it relates to me I can better detect where I've misunderstood.

Regards,

Truman
a c 177 V Motherboard
April 3, 2010 11:28:24 PM

RAID0, even in a FakeRAID, is decidedly faster than a single drive. I use two pair of RAID0 VeliciRaptors, in alternating system/swap volumes/partitions, and get roughly 200M/S reads - but I 'short-stroke' the drives to prevent fall-off toward the ends, using about two-thirds of the space, the very first partitions (at the 'nose' of the drives) being my normal 'working' system, seven U64, and on the other volume, its swap partition... I thought it would be similar to an actual RAID - the 'wider', the better; however, putting all four drives in one RAID array slowed it down by roughly 20%.

I assume SSDs would work the same - as SSD controllers are still not (even for SLCs) capable of 'saturating' the SATA2 interface spec...

Next planned upgrade (among other things): Areca 1680 4G 12 port card with 4 RAID0 Intel SLCs, and six WD REs for storage - still debating RAID 10 for those for speed, vs RAID6 for ultimate security - either way, 'lose' three of the six....

April 4, 2010 1:12:51 AM

Is there such a thing as a motherboard with TRUE hardware RAID? Like the way supermicro used to integrate the adaptec chipset?

And, to be clear, the 2-drive soft-RAID that I have on my Gigabyte WILL be faster than I single drive, yes? But you advice I DEFINE the starting sector or cylinder or something? Uhg! I thought that information was allowed to be forgotten (gleefully) when BIOS started auto-config drive size. :'( 
a c 177 V Motherboard
April 4, 2010 1:48:06 AM

If there is, I haven't come across it, but, then again, I haven't looked specifically for that, either... Lots of people do SiS chips on-board, for RAID SAS ports, but it's still mostly a FakeRAID deal... I predict that such a thing will be coming from Intel, however - QPI is oddly 'extensible'; at first I thought it was just for cache 'snooping' in multi-processor, multi-IOP, NUMA systems, like the twin 5520s, but, it seems to be a waste to me - I'm guessing we will see an extension of the IOP34x family, with a pair of XScale cores, handling disk I/O and parity calcs, while 'talking' QPI to a bus-ring of Xeons/5520s - it just seems like an inevitable target...
Quote:
And, to be clear, the 2-drive soft-RAID that I have on my Gigabyte WILL be faster than I single drive, yes?

Yes indeed!
Quote:
But you advice I DEFINE the starting sector or cylinder or something? Uhg! I thought that information was allowed to be forgotten (gleefully) when BIOS started auto-config drive size.

Not really define the start - you can just let the RAID BIOS take the whole disk, and partition the whole thing, too. But I've been doing this since the late eighties and windoze 2.0 - and I can't tell you how many 'sick' or 'dumped' windoze partitions I've seen :cry:  . I always advocate OS on one physical disk (perhaps serveral OS on several partitions on one disk), and data separate, on another drive! I still see scads of people who are 'timid' to properly 'fix' something by doing a reinstall - simply because they've 'left' their Documents folder on their install drive - and now they've got an intertwined MESS!!
The other point is that the beginning of a drive is the fastest part - and even the largest, extreme, seven ult 64 installs only need about a hundred meg - so I advocate using fast, small drives for the OS, and partitioning only what you need - your transfers are faster, and your 'seeks' are shorter!
April 4, 2010 1:40:29 PM

I agree with you with respect to OS and Data being separate, BUT, I do run in to problems pointing my special folders (My Docs, Desktop, etc.) at my D drive because, using 2 drives on a laptop means swapping out my data drive to put my BD-RW in to burn a disk - and voila, a problem emerges.

Anyway, I bought a cheapo RAID controller...

http://www.newegg.com/Product/Product.aspx?Item=N82E168...

It real-world transfered at any where from 90MB/s up to 140MB/s with 3 WD 500GB 7200 RPM drives configured in RAID-0. Individually, the drives get about 55MB/s... I haven't tried the FakeRAID yet, but, would you say that transfer rate is worth the $70 I paid for it? Ultimately, between the value of the drives and the card I could buy a 64GB SDD drive..... of course, I wind up with a tiny fraction the space.

I hesitate to say it, because as stated, I'm certainly outclassed, but I also do IT work -- and I'm shocked how often people want me to put a bandaid on their systems instead of a fresh install. They never realize how crippled their system has been since they received it with bloatware, let alone the 2+ years that have gone by since a clean installation. They have no idea the performance they're not getting, that they've paid for... and for MORE money they have me identify and resolve the problem then just fixing ALL problems with a re-install.

FYI, I'm setting this system up as an HTPC, and as of yesterday, I've discovered the goddammed COOLEST effing device that I've EVER seen for its problem;

http://blog.gsmarena.com/the-lenovo-multimedia-remote-w...

And when I say it worked INSTANTLY and perfectly, without an iota of setup, I mean it! The one thing I think it's missing (though I see an Fn key) .. is a way for me to hit Alt-Fn4, or the likes. That meager quirk aside, there's absolutely NO product that even CLOSE to performs as CONSISTENTLY, and in the form factor of a couch-surfer's usage-mode.

What do you recommend for RAID NAS? Preferably 4-slot, easy GUI interface with either native, intuitive, incremental and differential backup, user rights management, and decent transfer rates? Obviously, the only reason this is a difficult product for me to pick is because price IS a factor. :-)

OH! And final question... I have a Q9550 ... would you see any reason to switch to an i7 system? I personally can't find one.. and the performance difference relative to the cost of change seems like a poor value prop.

Okay.. :)  Talk to you soon. OH! I like your quote by the way.
a c 216 V Motherboard
April 5, 2010 1:52:16 AM

Quote:
I haven't tried the FakeRAID yet, but, would you say that transfer rate is worth the $70 I paid for it?
In RAID0, the ICH10R will be faster.
a c 177 V Motherboard
April 5, 2010 3:18:33 PM

GhislainG is right - very little (again, below the cost threshold where you get a separate processor) can even come close to competeing with the venerable ICH. Several reasons:

It is venerable [:bilbat:6] ! The ICHs have been through a pile of redesigns, while preserving the 'base' architecture; each iteration has taught Intel about how to do the physical access, as well as how to 'sub-optimize' the communications to the CPU. In addition, the drivers have been 'tweaked' and 're-tweaked' to get every last erg out of the hardware design...

Integration/'close-coupling' - as the CPUs are Intel's to begin with, no one knows better than Intel how to 'do the hook-up' better; in days gone by, several companied have tried to 'sue their way into' the chipset business, claiming Intel was hindering their ability to make a successful chipset product, but, think about it: you've got geeks wandering the halls at Intel who can tell you how many electrons there are in a 'saturated' QPI differential driver output stage transistor - it's just impossible to compete with that depth of understanding...

Bus transaction count - the ICH ports are directly on the southbridge chip, and talk to the CPU directly through the bridge bus; any other thing you can 'stick in' a slot, has to go through at least one, and sometimes more than one, 'layer' of 'translation' due to the slot location - PCIe, PCI, whatever; each one of these 'layers' is fraught with it's own costs, and it's own 'latencies'...

I wouldn't worry too much about the 9550 - I'm sitting in front of one dictationg here! I'm not happy with the 'maturation' of the 1366 platform boards, and will wait a few 'generations' before looking at 'desktop' boards; I think I already mentioned - my first 'jump' into i7-based equipment will likely be a server board - they can't afford to 'diddle around' with new, unproven hardware...
!