raid1 migration and EP45-DQ6 5723

[mv2devnull]

The EP45-DQ6 has two Silicon Image 5723 chips, each with two SATA ports and chance to run RAID 0 or 1. The manual naturally describes how to set up a new system.

But what if ...
Lets say there exists a pair of disks connected to Silicon Image 3114 chip on old motherboard, with RAID1 on them. Not system disks, so force feeding drivers to OS will not be an issue. But with data that preferably should transfer intact.

3114 is a standard fakeraid. 5723 boasts as hardware raid. Obviously, the data is already mirrored, so initialization is not necessary. Is there enough compatibility between 3114's raid metadata and 5723's format to pull this off at all? What to choose in BIOS?

BIOS on my DQ6 is version F5 at the moment. There seems to be couple newer version, but I do not like updates unless they are necessary.

[Lsdmeasap]

Re: raid1 migration and EP45-DQ6 5723

I do not believe you will be able to transfer the disks between the controllers. You should save yourself the headache and find some way to backup the data to a NON raid environment and then transfer it into your new board

[mv2devnull]

Re: raid1 migration and EP45-DQ6 5723

Who in sane state of mind would try such thing without backup? I think I'll test the migration once I have the backups.

[Lsdmeasap]

Re: raid1 migration and EP45-DQ6 5723

Ya, I just had to say that incase you did not plan to make some backups. Some people do not think ahead so I was just trying to help ya just incase.

Ya, I dont think it will work, but let us know how you get along with it

[mv2devnull]

Re: raid1 migration and EP45-DQ6 5723

It is good to be explicit. That benefits also other readers of threads.

Not quite there yet (no drives connected to GSATA ports). Noticed that the list of BIOS options in F5 is drastically shorter (and different) than the list in printed manual. Part of it could be due to "show only if there are drives", but not all.

Another observation:
Vista x64 boots nicely, but Linux kernels (CentOS 5.2, Fedora 9 pre-release) have timeouts while activating the GSATA ports (two, on JMicron). I have the GSATA on RAID mode, all four ports "normal". I should test disabling the GSATA entirely (but with IDE DVD cannot use that in "production"). Timeouts, resets fail, and port speeds drop to 1.5Gb. The same 'ahci' driver controls the six ICH10R SATA ports too, and they are fine.

I can Google similar errors, but not so far in same hardware setup. And solutions department is still lacking.

[Lsdmeasap]

Re: raid1 migration and EP45-DQ6 5723

Hmm it should be a different AHCI driver for each controller, or is that not what you meant?

1.5gb Eh, are you sure your Drives in that instance do not have jumpers on them? (One of those "Have to ask" kinda things again)

[mv2devnull]

Re: raid1 migration and EP45-DQ6 5723

Hmm it should be a different AHCI driver for each controller, or is that not what you meant?

1.5gb Eh, are you sure your Drives in that instance do not have jumpers on them? (One of those "Have to ask" kinda things again)

I'm sure, because there is no drives in there.

AHCI is supposedly a neat open standard, and therefore the 'ahci' Linux kernel driver is able to support multiple chips.


But to the wicked topic of the thread. (Btw, a rev 1004 of manual that I did download differs clearly from the printed copy in the box.)

So, I had in BIOS the Gigabyte SATA2 Chip 'Enabled', and 'Smart Backup Function' Disabled. No drives on GS0-3. PATA DVD on the IDE channel. During boot the ODD0 (the DVD) is found.

Now I took two Seagate 500 GB drives that had been a RAID1 on Sil3114 fakeraid, and connected them to GS0 and GS1.

On boot, the GSATA now found HDD0 in addition to ODD0, and offered the 'Ctrl-G' to start GRAID setup. The GRAID setup did show one 1000GB non-RAID disk of strange identifier (ie did not call it Seagate). I chose not to do anything there.

Instead, I returned to BIOS, and chose to enable 'Smart Backup Function' on GS0-GS1. Reboot and very quick "initialization" later I was back in GSATA, still seeing HDD0 and ODD0. GRAID setup had now one 500GB non-RAID disk. Time to go boot OS.

CentOS 5.2: the seventh SATA port was now seen as disk by 'ahci', with 3 Gbit/s connection. The port eight was still timing out as before (as there clearly is no physical device). /dev/sdc, a third hard drive was visible, and it was clearly partitioned. 'dmraid' did sniff a string that the Sil3114 did use, probably from metadata written to disk(s) by Sil3114, and called the drive "mirror", and "ok".

Vista x64 SP1: Sees the one drive as well. Offers "Safely Remove Hardware", a "SATA External SCSI Disk Device". And the partition with NTFS was automatically shown as drive "E:", with files apparently intact.


RAID1 migrated.


There was another thread recently, someone complaining that not all connected drives show up. I have a guess:

* The "Smart Backup Function" BIOS setting talks to the Sil5723 chips.
* The GSATA scanning drives is the Jmicron JMB363 (aka Gigabyte SATA2 chip).
* The GRAID setup is the Jmicron JMB363.
* JMB363 has one IDE channel and two SATA2 ports (and supports eSATA).
* One Sil5723 is connected to one SATA2 port.
* Sil5723 is by default in either JBOD or RAID0 mode. (Did Janus wrote about some BIOS rev that RAID0 is not there?) This explains the 1000GB "drive".
* In "Smart Backup" mode the Sil5723 is in RAID1 mode (or at least hopefully keeps copies coherent).
* Either way, the Sil5723 chip does show only one volume (an eSATA device) to the JMB363.
* There seems to be no way (BIOS nor application) to actually know the status of the "array" in the Sil5723.
* Even if the initialization of Smart Backup did copy everything (but it was too quick to do it, might work in background even now when I write this), it changes nothing, as the pair of disks was already a mirrored copy (differing only in the Sil3114 metadata). The only thing that might destroy data is if Sil5723 writes its metadata into the drives into bad spot. I do know from experience that Sil3114 and Nforce4 Nraid (both fakeraids) do write their metadata into different locations.
* The non-smart backup mode probably just concatenates volumes. Not sure, because I have no need to know.

Now, if I'm pedantic, I should disconnect one of the drives and see what transpires. That would test the existence of RAID1. But I have to wait until the "initialization" is surely complete.

What else have I forgotten or falsely assumed?

[Lsdmeasap]

Re: raid1 migration and EP45-DQ6 5723

Wow, Lotta info there. Hmmmm

So all went well then I take it. And what's this you say P45-DQ6 has no RAID 0 on the Gigabyte chip? If so I am glad you said something, I planned to get that board but now I may not

[mv2devnull]

Re: raid1 migration and EP45-DQ6 5723

I do not know whether it has RAID0 or not. If it has, it is the default mode. How one is supposed to interrogate the Sil5723 to confess what it is doing?

Disk benchmark probably will tell how good the default mode (that offers two X GB disks as one 2X GB volume) is. But I want RAID1, so I will not test that.

Btw, the GRAID setup does offer creation of striped or mirrored arrays, ie 0 and 1. But that must be on the JMB363, so it is fakeraid. Perhaps JMB363 RAID0 over two pairs of "Smart Backup" (presumed Sil5723 RAID1)?

But, this thread says (if I read correctly) that newer bios versions prevent the use of the JMB363 GRAID (probably to prevent people making RAID0 over RAID0).


Disclaimer: I know nothing, but guess fluently.

[Lsdmeasap]

Re: raid1 migration and EP45-DQ6 5723

Yes, you do guess pretty well. Sounds pretty bunk to me, Now I don't want one!

If I do get one, I think I will be modifying the BIOS and see if they just disabled the RAID0 options on the Silicon image chip from within the BIOS. If so, I would of course add that setting back. Then for sure shoot off a email asking why they would waste a Onboard processor for ONLY backup and not allow the use who pays 200+ for the board to choose how to use it

[mv2devnull]

Re: raid1 migration and EP45-DQ6 5723

It is more likely that they disabled the (fake)raid support of the jmb363 chip. At least on the F5 BIOS the 'Ctrl-G' setup is clearly configuring the jmb363, and the "sata2raid" utility from Gigabyte does the same from Windows. Now if you are told that you have (two) hardware raid chips, and then you are given bios tool and windows utility, which have nothing to do with the hardware chips, you will be confused, big time. Hiding those tools is a reasonable safety measure against trivial foolishness.

The "Smart Backup Function" is thus a toggle used to choose whether the Sil5723 chips do RAID0 or RAID1, and RAID0 (hopefully) is the default. What does worry me is that there is no option to choose a stripe size, as is customary for RAID0. Even the fakeraids offer that.

However, the Sil5723 should be able to do more than just RAID0 and RAID1, neat things like "LARGE", "SAFE33" and "SAFE50". So, in perfect world there would not be "Enable Smart Backup on GS0&GS1", there would be "Choose mode from: LARGE, RAID0, RAID1, SAFE33, SAFE50, ...".

If I had to bet, I would say that GA does not yet know how to let us configure the mode of the Sil5723 chips. After all, you have to talk to Sil5723 via the jmb363. And Gigabyte has to talk to Jmicron and Silicon Image.

[Lsdmeasap]

Re: raid1 migration and EP45-DQ6 5723

Well I know the J.Micron was the software based one. Now your next comments convinced me even more that They have advertised this board havign functions it dont.

I read you could use the Silicon Image for RAID0 and not have to use it as "Backup" only. Without being able to choose a stripe size that tells me right there the answer to my next question to you. Can you set it up for a NON Backup (OS) use, NOPE since you cannot even set the parameters Stripe or Chunk Sizes I doubt it can be used for anything useful to the large masses of users it appealed to when they bought it

Yeah, I dont think they do converse with them very well. A while back I was trying to get someone on the GA team to send me some P35 GA/J.Micron Drivers to be used in a DOS environment (16bit RAID Driver for Software development) Which they said they did not have, nor could they get. Asked J.Micron about it, and they sent me two variants one of which they said Gigabyte should have given me when I asked

I think all of our language barriers between everyone is the main trouble with a lot of software development and deployment for day to day use

[mv2devnull]

Re: raid1 migration and EP45-DQ6 5723

Look at Gigabyte and Silicon Image sites and:
* "Smart Backup" is the only advertised feature about DQ6/DS5 Sil5723 chips.
* 57xx SteelVine Manager Manual from Sil does not mention setting stripe, and it does describe the BIG, FAST (=RAID0), SAFE, SAFE33, SAFE50, and JBOD modes supported by the 5723 chip. So Silicon Image does not seem to include RAID0 parameter reconfiguration into their chip, nor mention what is their default (and only) stripe size.

EP45-DQ6 is expensive, but it includes two 5723 chips so you can have two "Smart" Backup volumes. A dual port hw raid card (Areca or 3Dware) costs about as much as EP45-DS5. Four port hw raid card is at least 1.5 times the price of EP45-DS5. Thus, if you buy cheaper motherboard and a 4-port hw card, you get two RAID1 arrays (or configurable RAID0), but you have to pay twice as much as for EP45-DS5.

Asus has one 5723 (behind Marvell rather than J.Micron) in some P5Q boards, but they are about as good as Gigabyte on this matter.

Bottomline: Sil5723 offers a cheap "hardware" RAID. Very cheap, very simple. "One button solution", if (and only if) the button does what you need. And if it does not, it does not.

[Lsdmeasap]

Re: raid1 migration and EP45-DQ6 5723

Ahh Well I had read only a user footnote about it really before it was available to the public. And Honestly I just seen hardware RAID chip and no mention of what it could be used for.

Guess that is what I get for assuming. I guess my new board will be P45-DS4 or P35-DQ6.

[mv2devnull]

Re: raid1 migration and EP45-DQ6 5723

Darn, I seem to have some read errors (according to SMART) in one of the drives I did use in testing. Have to sort that out first, before judging the SiI5723.