LSI RAID Controller help, 2308 chip on-board ASRock Extreme 11 motherboard is slow!

[Callsign_Vega]

Mobile, I flashed the IT firmware without a BIOS and it worked great. Fast boot times now and fast drives!

Any of you guys mess around with RAMDISKS at all? Wonder if it would be playing around with putting games on a RAMDRIVE or if it would not be that much of in improvement over SSD's.

[AikenDrum]

Do they still make zero-channel raid controller cards? The ones that direct traffic, but don't actually own the drives/ports. I don't know if that would even be a good idea for this, let alone feasible, but I'm curious.

The reason why I'm curious is that there is (or was at the time) so little on a ZCR that the profile is incredibly low and narrow and you might very well be able to fit it on one of the little 1x slots between the 16x slots that we want to put four graphics cards in.

What I don't know/remember about ZCR, since I never actually used one and I'm really quite ignorant about it, is whether you need motherboard support for it, or what sort of performance it can actually give. I don't think it needs to handle the actual data being passed around--it just needs to direct it. If it needs to pass the data through its own piddly little connection, then the 1x speed (and 2.0 as I recall) is not gonna cut it.

Just thinking out loud. If I'm being a moron, please disregard.

[mobilenvidia]

The LSI HBA's have no cache, but are fully fledges SAS2/SATA3 controllers
As we've proven up further is that HBA's run best when in IT mode where the drives are just presented to the OS.
Letting the OS do any RAID if needed gives best performance.

[Computurd]

well here is my write-up, even though i try not to link my own drivel

9207-8i

[mobilenvidia]

Nice writeup Paul, nothing wrong with blowing ones own trumpet

One thing though, is the SAS2308 a dual core or a faster clocked SAS2008 (single core) ?
LSI mentions nothing of dual core, some other places do mention dual core.
I'm leaning towards a faster SAS2008, just sngle core.
As the performance increase looks to be the 533Mhz to 800Mhz increase and PCIe Gen2 to Gen3 latency decrease

[Callsign_Vega]

well here is my write-up, even though i try not to link my own drivel

9207-8i

Nice. I would have like to have seen an AS-SSD screenshot of the 8x C400 setup to compare against my setup but no biggie. Really, all of these HBA's need to be run in IT mode eh to get the best performance.

[Nizzen]

Can't get my 9207-8i retail to work properly with EVGA 1155 FTW Is it impossible for EVGA to get sascontrollers to work with they're motherboards?

[Snuddi]

Can't get my 9207-8i retail to work properly with EVGA 1155 FTW Is it impossible for EVGA to get sascontrollers to work with they're motherboards?

My Adaptec 6805e worked flawlessly on Evga Classy X79, so they can do it.

[tived]

how do you do that?
asign them as workers/manager...

please
thanks

henrik

for even higher speeds...do not RAID them with winders RAID. it is no good for scaling. configure them as separate volumes, then assign either a worker, or manager, to each. Then you will see top speed. no raid!

they have actually refined the 2308 and FW to where they can pull 700K IOPS now, i reached around 633K and my hardware was the limitation.

[Greg83]

I've messed around the with ramdisk a bit. this loads at bootup 22GB , with my games, users folders and apps. to keep everything smooth.
I disabled write caching here. thought why bother. lowered as-ssd scores but , meh its fast. This is just ram running a basic 1212Mhz. quad channel might kick major ass.
Using Primo Ramdisk Ultimate
http://i49.tinypic.com/11gm51g.png
(couldn't find how to shirk the image for here.

[Nizzen]

LSI 9207-8i HBA
7x Plextor M3p 128gb + 1x intel 520 128gb ( waiting for last m3p :p). Software raid-0

Asus x79 WS 2011, server 2008r2



7x m3p software r0: Copy bech


Atto 7x m3p + 1x520

So close

Edit:
QD=10 dit it

And there is no cache what so ever

[Callsign_Vega]

I wonder why the Seq write speed is so low? Does the M3 Pro's not like AS-SSD's incompressible write test?

[Nizzen]

I wonder why the Seq write speed is so low? Does the M3 Pro's not like AS-SSD's incompressible write test?

M3p does like incomppressible write, so I think it is lack of Areca firmware

I use the "it" june firmware. Have not flashed to the july firmware yet.

[Callsign_Vega]

I have 64 GB of 2400MHz G.Skill arriving tomorrow. Should running pretty fast in quad-channel RAM-Disk.

Got my Plextor M3Pro OS drive working properly. I used the latest 11.5 and it said "this driver isn't guaranteed to work blah blah" and I forced it to install. Now the numbers are great:



No one should be using the C600 X79 Intel drivers those things are horrible.


Even though the controller is unrelated, broke a 4k score on my 8x Vertex 4 array:



Games on that array should load pretty quick onto the RAMDisk.

[Andreas]

Mobile, I flashed the IT firmware without a BIOS and it worked great. Fast boot times now and fast drives!

Good to read you found the culprit.

Any of you guys mess around with RAMDISKS at all? Wonder if it would be playing around with putting games on a RAMDRIVE or if it would not be that much of in improvement over SSD's.

No. Not since the days of MS-DOS (Remember? Last millenium).
With smaller SSD RAIDs the RAMDrive would show some perf benefits. If you take multiple LSI controllers in concert, the RAM Drive develops a "weakness" vs. the SSDs. Utilizing the RAMDrive at higher speeds, the mem bus will be "loaded" twice when transfering data from the ram drive into main memory. This leaves less bandwidth for your application.
Another thing with RAM disk is a higher CPU load vs. a good IO system. Usually the ramdisk copy is done via the CPU and not the DMA subsystem.

I am running the LSI controllers at 14 GB/sec on a stock X79 mobo, leaving still 16 GB/sec of remaining mem bandwidth with 1333 DDR Ram (24 GB/sec with 1600 DDR3). A ramdisk would need 28 GB/sec for similar performance (seen from the app). Leaving only 2 GB/sec respectively 10 GB/sec for the 3930K CPU. Pretty unbalanced system.
IO transfer at this speed needs less than 5% CPU time on a stock i7-3930K CPU. Haven't checked it, but I assume a RAMdisk at this speed will consume 1-2 cores fully.

I have 64 GB of 2400MHz G.Skill arriving tomorrow. Should running pretty fast in quad-channel RAM-Disk.

Rather make sure that the RAMDisk Software drives the RAM chips effectively (for real world performance).
Easy test:
Run the stream triad benchmark on your idle system.
install ramdisk sw, start a benchmark with max io bandwidth (i.e. iometer) - keep the io benchmark running.
start the stream triad benchmark again (while iom is running).
What are the new numbers for IOM and stream?

No one should be using the C600 X79 Intel drivers those things are horrible.

They are fine, as long as the DMI 2.0 link to the CPU isn't oversubscribed. It's only 2 GB/s peak (1,6 GB/sec practical)
On your ASRock MB a fully loaded IO section ( 8xUSB3.0, 14xUSB2.0, 1xFW, 2xeSATA, 2xGbLAN, 2x6GbSATA, 4x3GbSATA) would provide/consume approx. 78 GBit/sec to be delivered over a 20 GBit/sec DMI backbone - it's 1:4 oversubscribed. Seen the other way around - 4 x USB 3.0 connected SSDs would saturate the link, leaving limited bandwidth for the SATA ports.

Even though the controller is unrelated, broke a 4k score on my 8x Vertex 4 array:
Games on that array should load pretty quick onto the RAMDisk.

Nice number.

BTW, the Vertex 4 need regular TRIM support to maintain their fresh performance. As soon as you leave performance mode and enter storage mode - write perf drops ca. 75%. Can't be done via RAID/HBA adapters these days (except Intel's RST). Need to be done on one of the X79 SATA ports. No big issue, just something to consider.


Have fun with your system,
Andy

[Denonic]

Hi guys,

Sorry to post in a fairly old thread but most of my issues seem to have a lot of info here.

I recently came back from a holiday in Hong Kong, and having an Extreme11 and being quite under budget I decided to grab some SuperSSpeed S301 64GB drives to test on the LSI controller.

My problems will be easier to see with some screens:



So the issue seems to be somewhere with my LSI controller because the Intel MSAHCI 4k-64Thrd numbers are decent.

Here is an ATTO benchmark that similarly show the issue:



I don't know what I could possibly be missing or if the differentiation between Intel and LSI ports should be that much.

That's the biggest problem I'm having.

Mobile, I flashed the IT firmware without a BIOS and it worked great. Fast boot times now and fast drives!

I've also flashed to the IT firmware without flashing the BIOS but my boot times still increased by 20 seconds.

If you are going to run the SAS2308 in IT mode, flash the Firmware but don't flash the BIOS.
This will speed up boot time no end as it won't load the BIOS as there won't be any.
The drives will still show to the OS (JBOD) as normal.
You just won't be able to boot to the LSI BIOS, which in IT mode there nothing to see or do anyway.
To do this you will need to cleanflash the SAS2308 first then load only the Firmware.

Is there an extra step required to cleanflash?

The steps I followed were:

sas2flsh -o -e 6
sas2flsh -o -f <firmware>

I have seen instructions that say to do this:

megarec -writesbr 0 sbrempty.bin
megarec -cleanflash 0
<reboot, back to USB stick>
sas2flsh -o -f <firmware> -b <bios>
sas2flsh -o -sasadd 500605b0xxxxxxxx (x= numbers for SAS address)
<reboot>

But I'm not so sure that I should be rebooting during the flashing process.

Any advice would be greatly appreciated.

Thanks


Miguel

[mobilenvidia]

Megarec will most likely not even detect the SAS2308 controller, it's for SAS2008/2108 controllers

In IT or IR mode you also don't need to do anything with SBR as this is only for IBM M1015 (and other OEM versions) of the LSI9240

sas2flsh -o -e 6
sas2flsh -o -f <firmware>

Should do it, BUT BECAREFUL !! don't reboot as you may end up with a paperweight.
I would really avoid this, unless absolutely necessary or you have money to burn.

You would be much better off just flashing the card with IT Firmware, just do it with MSM in windows or sas2flsh -o -f ,firmware> or megacli
You only really need to clean flash when you go from IT to IR (or otherway)

You never flash the BIOS in IT mode as the whole point of IT mode is not to have a BIOS
Do how ever flash the UEFI BIOS/Driver, this allows quick boots for system with UEFI BIOS.

[Denonic]

Megarec will most likely not even detect the SAS2308 controller, it's for SAS2008/2108 controllers

In IT or IR mode you also don't need to do anything with SBR as this is only for IBM M1015 (and other OEM versions) of the LSI9240

sas2flsh -o -e 6
sas2flsh -o -f <firmware>

Should do it, BUT BECAREFUL !! don't reboot as you may end up with a paperweight.

You would be much better off just flashing the card with IT Firmware, just do it with MSM in windows or sas2flsh -o -f ,firmware>
You only really need to clean flash when you go from IT to IR (or otherway)

You never flash the BIOS in IT mode as the whole point of IT mode is not to have a BIOS
Do how ever flash the UEFI BIOS/Driver, this allows quick boots for system with UEFI BIOS.

Alright that was what I did.

Now I just need to figure out why the 4k-64Thrd writes are so slow.

[mobilenvidia]

Have you played with the Drive cache setting in Device Manager, Disk drives ?
Reboot each time you adjust

[Denonic]

yep. i get a boost of 250-300MB/s when i tick turn off windows write-cache buffer flushing on the entire array so 800-900MB/s. Still a far cry from what it could be considering the Intel port had about 300MB/s 4K-64Thrd writes for a single drive. There is no option for removal policy though so not sure if that makes any difference?

[Denonic]

So I just went and did a whole lot of benchmarks for a review on my blog and ran into something interesting that could explain the low 4k reads on the LSI ports of this board. I'm unsure if its isolated to mine but this is something I was able to compare between the Intel ports and had interesting results:



LSI


Intel



LSI


Intel



LSI


Intel


Anybody have any theories as to what could be causing that pretty substantial dip at the beginning of the graph?