Topic: How do I do hot-swap SATA?

[darkgumby]

I have a client running 6.0.
I have 2 SATA drives in the system.
The primary drive has SME installed and it runs fine.
I want to use the secondary drive as a hotswap removable backup drive.
I have an enclosure that is supposed to be hotswappable.
I can mount and umount the drive with no problem. If I umount the drive and then unlock the enclosure the whole system freezes.
The enclosure is supposed to be hotswappable so I doubt that's the problem.
I am using a Silicon Image 3112-based SATA controller card.
What can I do?

Does anybody do hotswap SATA with SME? If so then please describe your setup and how you do hotswap?

[dmac]

Haven't done Hotswap with SME yet.

Question would be whether your SATA chipset supports Hotswapping?

[mbachmann]

Right, the chipset must support hotswapping and the sil 3112 doesn't. An enclosure can never support hot swapping or online rebuilding or whatever.

Those controllers are not found below a 150$.

[Reinhold]

Hi

mbachmann is right about the lack of capabilities of the sil 3112 in this respect (or in general if you ask me .-). It's even "funny" that it works as SATA at all in SME !?
However FWIW there are a number of "hot swappable boxes" out there ... several for standard ATA. I believe Darell May even sold one type for a while.
Basically those behave like a single drive to the system doing everything internally, mainly RAID1 (10 maybe 5) hot swapping and rebuilding included.

darkgumby: too little info on your system given to really reply but did you use raidhotremove and raidhotadd as well?

Regards
Reinhold

[arrowsmith]

I have tried to hot swap a SATA once on windows XP.  It froze up on that aswell.

i would generally not recommend to do it unless you have a SATA hot swappable rack.

[dmac]

You don't need a hotswap rack, most of the hot swap racks that I have investigated all indicate that the Chipset needs to allow hotswapping.

It is quite possible that the chipset you are using does not support Hot Swapping.  The best combination would be a good SATA Raid card (I recommend the Promise Fasttrak S150 SX4) and a hot swap rack, but you can use any removable rack with a good SATA raid card.  I have recently setup a Windows Server with a Gigabyte Motherboard that has onboard SATA Raid and this chipset support hot swapping.  You can just pull the drive out and it does not effect the operation of the OS.

Check out the specs on your SATA Raid chip set and you will most likely find it does not support Hot Swapping.

[Reinhold]

dmac,

I do have problems with your last post ----->

<sarcasm>
So you do recommend (!) to us SME'ers that crappy, slow controller called S150 SX4
that has problems with most type of DRAM and afaik is, as of 12/2004,
without _any_ usable drivers for SME ...

GREAT - offer accepted, give us your postal address...
A friend will send you his S150 SX4 he bought after a recommendation like yours.
Will you pls perform some magic and it will run faster than SME built in software raid and <tada>
even hotswap without a removable rack in SME ?

I will even ignore: http://linux.yyz.us/sata/faq-sata-raid.html#sx4

  </sarcasm>

go... investigate... can only hope you mistook this as an XP thread  

Reinhold  

[dalex]

Ok hot swap is an expensive option...

How about COLD swap?

This is not a very clear area for me. I have all my installations as ATA and SATA raids (SME in software raid 1) thinking i am safe. In my 1st (ata) disk failure (thanks god i were looking often the logs - there is no other way in SME to find a disaster is coming...), the wrong disk were the 2nd/slave(?). So i shut down, changed it with a brand new, rebooted, and after a couple of hours (server working!) i had a new healthy mirror.

Te 2nd time i had no luck... The faulty disk were the 1st/master. When i rebooted the system hanged! No boot record found! So, is the mirroring done ONLY in the DATA area (partitions)? How about the boot sector? Anybody knows? Boot diskette is the ONLY solution? There is no warning in the installation procedure...

[dmac]

So you do recommend (!) to us SME'ers that crappy, slow controller called S150 SX4
that has problems with most type of DRAM and afaik is, as of 12/2004,
without _any_ usable drivers for SME ...

GREAT - offer accepted, give us your postal address...
A friend will send you his S150 SX4 he bought after a recommendation like yours.
Will you pls perform some magic and it will run faster than SME built in software raid and <tada>
even hotswap without a removable rack in SME ?

I will even ignore: http://linux.yyz.us/sata/faq-sata-raid.html#sx4

No reason to get nasty.  I should clarify that I would recommend the Fastrack S150 SX4 with a Windows Server.  If you read the entire post you would have noticed the reference to the Windows Server.  I agree with your rant that the linux drivers for the Promise line of Raid controllers is a bit limited.  I tried as many have to compile the drivers but my experience has been that the drivers are written for a latter version of the C code and Kernel than the SME server is currently running.

I checked your link, and would disagree with the comment made that the SX4 is not a hardware raid controller.  When the SX4 is installed, the raid card handles the raid array, the OS only sees the drive or drives array that have been configured.  When the drive fails or is removed and replaced, the card handles the rebuilding of the array, with no impact on the OS.

Hopefully when the SME server is built with the new Fedora Core, Hardware driver will work a better.  Incedently , using Fedora Core 2 this

crappy, slow controller called S150 SX4

works well.

Darin

[dmac]

dalex,

Te 2nd time i had no luck... The faulty disk were the 1st/master. When i rebooted the system hanged! No boot record found! So, is the mirroring done ONLY in the DATA area (partitions)? How about the boot sector? Anybody knows? Boot diskette is the ONLY solution? There is no warning in the installation procedure...

Did you switch the drives around and make the secondary the primary?  If i'm not mistaken, the Software raid will mirror the entire drive, including boot sectors, but since your new first drive does not have the boot sector it will fail.

[Reinhold]

dmac,

Yes it's a rant ... but with some   ...and pls note this thread was about "a client running 6.0"

Please don't confuse SME users as to the facts:
WARNING - the Promise S150 SX4 doesn't work in SME as of 12/2004 - nobody can make it work because of lacking driver/kernel ! - WARNING

It seems to (sort of work) with Feodora and Suse 9.x ...but  
i) the SataFaq author is right (definitely!) I doubt even XOR works in Linux...
ii)have you taken any data? In my friends 2.8 Ghz Server even when it (sort of) works it's slower than a Linux software Raid1
and we tested about 10 SDRAM sets before one would work.
(they even need to include a special program for that test! <lol>)

For a few Euros more you can get a 3ware if you want working hardware.

Regards
Reinhold

BTW: I do run/own support about 16 Promise controllers  - the "IDE-only"'s are good the rest is "cheap"

########## addition ##########
As I strongly believe in data not rants <eg> ... I had the other proud owner in a chat a few minutes ago... this is some simple data from the friends "fully supported S150 SX4 in Feodora" ...
-----
# hdparm -tT /dev/md0

/dev/md0:
 Timing cached reads:   660 MB in  2.00 seconds = 330.05 MB/sec
 Timing buffered disk reads:   48 MB in  3.11 seconds =  15.43 MB/sec
-----
He said he had a measured write rate to the fs of about 7 MB/s ... which is slower that a 100MBit line

----- and this is a software-raid5 (!) 1.8GHz AMD gateway I am sitting at right now
-----

# hdparm -tT /dev/md3

/dev/md3:
 Timing buffer-cache reads:   128 MB in  0.51 seconds =250.98 MB/sec
 Timing buffered disk reads:  64 MB in  1.03 seconds = 62.14 MB/sec
---- and that's a drive itself/alone ---
# hdparm -tT /dev/hda

/dev/hda:
 Timing buffer-cache reads:   128 MB in  0.51 seconds =250.98 MB/sec
 Timing buffered disk reads:  64 MB in  1.10 seconds = 58.18 MB/sec

Note: The hard disks are of course different and this is a weak hint type test BUT I'd be really interested in your S150 SX FEODORA DISK DATA...

... @darkgumby Sorry I am starting to hijack your thread here <smile> ...

[Reinhold]

dalex,

"Cold swap" is usually not a real problem:
- deactivate the dead hd (remove from raid by command or raidtab)
- shutdown
- turn off / exchange drive reboot ... fdisk&mkraid
...look for the raidmonitor stuff around here or read the RaidFaq...

Raid1 (definitely the SME Raid1) does copy everything (just look at the /etc/raidtab what is raided) incl. boot and root...

The reason a setup won't boot is most likely that the 1st DRIVE FAILED and the Bios that still wan'ts to boot from DRIVE1  (hha or IDE0:0) - go into the BIOS to fix that.

Regards
Reinhold

[dmac]

Reinhold

I realize the original post was based on the SME 6 release.  But it has also turned into a bit of a general discussion on Raid cards and Hot Swapping.  

The Fedora Core configuration was a test with the Promise Card when I had it in my possession. It has since moved into a Windows Server box.  If memory serves me correctly, It was a basic Fedora installation with the Promise RAID driver build after the install.  The drives attached where 200GB Seagate Baracudas and I must admit, I didn't collect a lot of data on the speeds the system was running at.  The configuration was using a IDE as boot with the RAID card setup to host the redirected /home and /usr folders.

[dalex]

Thank you all for the enlightening answers!