Sata Port Multiplexing RAID1 Data from Command Based Switching port multiplexer

GenesisFactor

[H]ard|Gawd
Joined
Jan 22, 2007
Messages
1,262
I just hope someone finds this helpful.

Overview:
This post just shows the performance of a Command Based Switching (CBS) SATA port multiplier/multiplexer (PM) using a software based RAID 1. The GA-D525TUD has an onboard JMB363 sata controller which uses the older CBS based instruction for port multiplexing. This allows only one drive to be accessed at a time. The newer, or better controllers, use FIS/FBS, which allows multiple drives to be accessed. A better explanation is here. I had searched on line for someone else trying this out, and I found no resources, so i figured i would just post it up. While I STRONGLY suggest that you get board or add on card with a FIS/FBS based PM, for some, that might not be an option.​

Data (Because no one really cares about the back story anyways :) )

Drives- 2x 500 Hitatchi Ultrastar A7K2000 (thanks dbwillis)
Enclosures- 2x Sans Digital TR4M+BNC
MB- GA-D525TUD
Chipset- Onboard JMB363 sata chipset using Command Based Switching Multiplexer.
OS- Ubuntu 11.10
RAID- mdadm Ubuntu Software RAID (you can do PM or RAID on the JMB363, not both)
PSU- PicoPSU


The although i did the 400GB drives (WD RE4's, which had TERRIBLE comparative performance), 500GB drives was where i truly experimented

Drive 1
500GB2singledrive-1.jpg


Drive 2
500GB1singledrive-1.jpg


Drive in RAID1 on Same TR4M
500GBBMonsameTR4M-1.jpg


Drives in RAID 1 on Same TR4M while being accessed
500GBBMonsameTR4M-duringaccess-1.jpg


Drive in RAID1 on Seperate TR4M
500GBBMonseperateTR4M-1.jpg

Some Thoughts....
The drives worked as expected in each case. There were nice read and write speeds for each drive as a single drive, . The write speed in RAID1 vs single drive practically halved when using the same case, they the read speed remained the same, as expected. There is a 17.32% (71.0MB/s vs 58.MB/s) write speed boost using it in separate cases on separate ports. The read speed is pretty much the same at 79.9MB/s. I believe that something accessed the 400GB in the second case, giving that ultra low 313.9 kB/s. With CBS, one drive can be accessed at a time- bummer. This point is further reinforced with the drives being accessed chart. I will post up one more with a drive being benchmarked while another is being accessed. This might not be enough to justify buying a second case for many. A second case would be good to house each array separately, as i can only access data one drive at a time. I probably wouldn't think of doing RAID5 off of CBS, so this is the best redundancy you will get.

If you are running two arrays at once, and foresee them both being used concurrently for sustained periods, you will need two enclosures with a RAID in each, else CBS is not the way to go. You will get staggered instructions and transfer times. For random periods, you can probably get away with it. Most add on cards, such as the RocketRaid 622, use FIS/FBS, so hopefully you will go that route. Just remember to READ THE SPECS.

Port Multiplexing is pretty awesome though, and GREAT for tiny system builds with external storage (like an HTPC). I personally prefer ESata over other interfaces, although its certain that USB3.0 will take the day as the preferred method of external connections. Hopefully that will not deter manufacturers from putting PM capable chipsets in their boards. On a "green" note, these drives were previously in an E-ATX case, and that system ate up almost 500W (actual), and would suck as a file server. This system configuration uses up less than 1/3 of that.​

Background
I upgraded my M-ITX server, begrudgingly, to the GA-D525TUD motherboard instead of an APU based motherboard, despite my desire for virtualization due to the price and my need for PCI expansion for my SCSI. Looking back, the mixed bag turned to my favor, as Hurricane Sandy rolled in a few weeks later and my UPS provided ample power for more than a few hours with the set up.

One of the things that the GA-D525 has is the Jmicro JMB363 sata controller, with sata port multiplexing. For the most part, this was a superfluous feature, but after Newegg's sale on the Sans Digital TR4M+BNC for $55 each after rebate, i had to jump in and get a couple just to see how well it works. Up until now, my board had run caseless, as i had connected many drives, but didn't have the funds to justify a new case. This new arrangement allows me to use my mini ITX case, as most of my drives now exist outside of my file server.
 
Last edited:
As an Amazon Associate, HardForum may earn from qualifying purchases.
One more thought. It seems that most people DIDN'T EVEN KNOW the GA-D525TUD has this capability. A few regarded guides actually have you add a port multiplexer adapter or a sata controller card to your build.

http://senk9.wordpress.com/2011/01/15/diy-home-nas-build-part-1-hardware/

and

http://www.overclockers.com.au/article.php?id=918715&P=1

The latter has 7 drives in the install. I currently have 9 with space for 2 more (1x IDE) if you include the DVD burner.

I wonder how many other motherboards are out there that failed to document PM as a feature?
 
When I was chasing PM options a couple years ago I found information in Intel documentation describing FIS PM functionality being available in all their SATA controllers as of, I want to say ICH7.

I never found driver software to make it available in the OS for my project and 6G was becoming common so I moved on. But yeah, it is not widely known that PM capabilities are baked in.
 
One thing to keep in mind regarding PM (or any other supported characteristics) is just because a chipset/controller in general is capable of a particular functionality is no guarantee that any particular implementation has enabled said functionality.
 
One thing to keep in mind regarding PM (or any other supported characteristics) is just because a chipset/controller in general is capable of a particular functionality is no guarantee that any particular implementation has enabled said functionality.
True. The LSI chip supports Ethernet in their schematic but Areca is the only maker I know of that enabled it and gives it an external port. :(
 
I see your points, mwroobel and vr. Thank you for clarifying. :)

One thing to keep in mind regarding PM (or any other supported characteristics) is just because a chipset/controller in general is capable of a particular functionality is no guarantee that any particular implementation has enabled said functionality.

Interesting, But isn't that decided on the driver level? (this is me being inquisitive, so help a brothah refine his understanding :)). My *basic* understanding of computer/electrical engineering is that some functions are circuit based, while others are on die based. the above case, i've seen before, and I can get a physical header or lack of circuitry, but i would have figured that the fact that the sata interface existed, most of the high end features also would. I'll have to see the functional circuit diagram for the Jmicron chip, but could such a feature, due to the complexities, speed, and required expected characteristics could possibly make it a System on Chip? I mean, even the Addonics card is a very simple looking circuit.

Also, because I'm running Ubuntu, and Jmicron provides full drivers and compatibility for all their chipsets to linux kernels, could it be that Gigabyte doesn't have drivers for this feature, a'la macbooks that had wireless N cards, but had to pay for new drivers to enable that feature? I never did a windows install on this machine, so it's possible, although i REALLY wanted to have virtualization (why i went for the APU originally)

Thanks for your help in clarifying...this is really messing with my head why manufacturers will have a working technology and not tell anyone!

@ vr, yeah i do remember reading about that with intel. They had some partial implementation for their own inhouse purposes (rapid storage technology, etc). That is definitely something to look into. I think 3rd party controllers baked onto their motherboards are the only real way to get it, What sucks is that i've looked through several motherboards, and the only documentation on what chipset you get with it are *usually* in the NDAs. In fact, i think only one had the chipset on it, and it wasn't the complete name, which is understandable for revisions. I guess the only way to check is to buy it and install (that's how i tumbled across it) and research from there...WHICH BLOWS.
 
Last edited:
Back
Top