Replacing SCSI controller card, need opinions, not a desktop

F

feigned

[H]ard|Gawd
Joined
Jan 7, 2003
Messages
1,224
Looking to replace our aging Adaptec 3210S controller card today...going to overnight it from Japan and put it on the street where it can make me money.

But seriously, it took a shit today and I don't feel like dealing with it. Pretty sure it's out of warranty, too. I don't really care if it is or not.

Just looking for some options. 3Ware, Tekram, LSI Logic, et cetera. Would like battery backup but it is not necessary.

Dual channel, U160/320. My HBA is U160 so a 320 card isn't entirely necessary. RAID is a necessity however. Gobs of cache if possible. PCI-X form factor, I can do 66MHz without an issue, not sure about 100 or 133.
 
The card linked is okay, but the price is too high. Try the 320X2128 from Newegg for $600 instead of the (older!) 320-2 for $750. And as an added bonus, you get a card based on the IOP321 instead of the IOP303 for four times the frequency -> better raid 5 performance. It has 128MB cache included, and a battery backup/expansion to 256MB is available, part number LSITBBU03. It's about $240 from what Froogle tells me, if you can find stock anywhere :( Maybe buy the card now and add the BBU later?

Intel also makes a controller based on the 321, but I'd be inclined to buy LSI. I'm sure the enterprise-type people will stop by and give their input.

 
unhappy_mage said:
The card linked is okay, but the price is too high. Try the 320X2128 from Newegg for $600 instead of the (older!) 320-2 for $750. And as an added bonus, you get a card based on the IOP321 instead of the IOP303 for four times the frequency -> better raid 5 performance. It has 128MB cache included, and a battery backup/expansion to 256MB is available, part number LSITBBU03. It's about $240 from what Froogle tells me, if you can find stock anywhere :( Maybe buy the card now and add the BBU later?

Intel also makes a controller based on the 321, but I'd be inclined to buy LSI. I'm sure the enterprise-type people will stop by and give their input.

Click to expand...
Wonderful information...this is why I posted here. Looks like a solid option, but I won't be ordering until this afternoon so...

Mainly looking for RAID 1 performance over RAID 5 as the controller will contain 5 mirror sets.

Isn't Areca a badass brand to go with as well? Or are they more on the SATA side of things?
 
Raid 1? Pfft, anything can do that. If you might someday go to a raid 5 array, then stick with that card; otherwise save a bunch of money and get this.

Hmm, on second thought, that has no onboard cache. If you need cache (and it sounds like it wouldn't hurt) stick with the one I suggested.

BTW, 3ware and Areca don't make scsi raid controllers. Tekram does, but they appear to be low-budget crap. I haven't tested them, but I would stay away until I hear otherwise.

 
unhappy_mage said:
Raid 1? Pfft, anything can do that. If you might someday go to a raid 5 array, then stick with that card; otherwise save a bunch of money and get this.

Hmm, on second thought, that has no onboard cache. If you need cache (and it sounds like it wouldn't hurt) stick with the one I suggested.

BTW, 3ware and Areca don't make scsi raid controllers. Tekram does, but they appear to be low-budget crap. I haven't tested them, but I would stay away until I hear otherwise.

Click to expand...
Yeah, Progress database with those five mirrors. Would really like to have the cache for that reason, unless it makes no difference in that environment.

Just called LSI and I'm having them e-mail over a compatibity sheet for the server that it would live in. Shouldn't be a problem but I wanted to see if they had tested it against the same hardware and if any issues had came up. I don't think it will be a problem otherwise.
 
Hmm, just thought of something - don't trust what's on the disks to work with the new card - different manufacturers have different on-disk formats, so you could get corrupt data. Wipe and restore from backups.

Please

Please

Please tell me you have backups :eek:

Any reason your databases aren't on SAME (stripe and mirror everything)? It's a very popular way to balance loads across several databases and disks. Use a large chunk size (8MB) so as few disks as possible are involved in any one transaction, and you end up with something about like load-balancing queueing for disks, with very little work on your part. Are you the DBA, or could you ask them about this? If performance is an issue (and I'd guess so ;)) it could help. Also see the KB article on performance (I assume it's the same Progress database...):
The most effective way to balance the load across multiple drives is to create a stripe set that combines all the drives into one logical drive with the data evenly spread across them. But if you lose one drive, you lose all so you have to combine striping with mirroring to get reliability.
Click to expand...
They agree with me :D I feel so loved...

 
unhappy_mage said:
Hmm, just thought of something - don't trust what's on the disks to work with the new card - different manufacturers have different on-disk formats, so you could get corrupt data. Wipe and restore from backups.

Please

Please

Please tell me you have backups :eek:

Any reason your databases aren't on SAME (stripe and mirror everything)? It's a very popular way to balance loads across several databases and disks. Use a large chunk size (8MB) so as few disks as possible are involved in any one transaction, and you end up with something about like load-balancing queueing for disks, with very little work on your part. Are you the DBA, or could you ask them about this? If performance is an issue (and I'd guess so ;)) it could help. Also see the KB article on performance (I assume it's the same Progress database...):

They agree with me :D I feel so loved...

Click to expand...
Should be fine just breaking the mirrors then creating new arrays. I wasn't going to plug and pray it. :p

And of course I have backups. The admin before me didn't back up when we were moving to this platform from a dual P3 system...disks were dynamic...I have a thread on "that other" forum if you're interested in seeing it.

And the only reason why I'm not running 0+1 is a reason that is listed in the link in your last post. The other one is that I kind of fell into this position. At this moment I don't really care. Although, if I'm going to rebuild arrays anyway, I guess it wouldn't be that much of a stretch to get a new array setup going. but it's been on my mind...back burner project that may have to see the light soon.

2. Stripe data extents on as many separate spindles as possible.
Click to expand...
Our extents are spread across three different mirrors. Hence the last paragraph I wrote out.

16. Do not run other stuff on the database server machine.
Click to expand...
This is the one that absolutely burns my fucking ass. The other admin before me decided to keep the machine a domain controller and not leave it to do db only stuff. His response: "It's powerful enough, it can handle it." More than mildly irritating. Oh, same response for the Exchange server too...and guess what it is besides serving up mail? :rolleyes:
 
All right, ordered the 320-2x without the BBU, but getting a 256MB ECC Reg stick of SDRAM to slap on the thing.

Not going to 0+1 any disks yet, want to talk to Progress before I go and dork everything up. :p

Thanks for the help mage.

avatard.gif
 
Installed the card today. Went great after I figured out that the blanks the drives after you initialize the logical disks. Ended up having to recover one drive's worth of data over the network using GetDataBack for NTFS.

Tried to perform a backup of the damned thing beforehand but it just wouldn't take. I should have been a bit more careful but in the end there was no data loss. The database lives!

Haven't installed the 512MB of DDR ECC memory for the controller yet either. Nor the rounded single end SCSI cables I ordered. I'm kind of miffed about that...the SPSH4 chassis we have absolutely 100% sucks for the 5 device cables. Routing those isn't nearly as easy as IDE cables.
 
Glad it worked out! Good luck with the RAM and the cables. Did you decide to go with the raid 1 arrays or one big 0+1?

Oh, and since you didn't get backups before... get 'em now!

 
Spoke too soon!

Ugh. 2nd worst week ever so far.

Got a call four hours ago so I'm back in the office wearing hole-y Levi's, a grey shirt that says "For Sale" in green letters, and my sandals. If they don't like it they can kiss my unshaven ass, facial cheeks, and my unkempt hair.

The card apparently decided it was time to lose an array and spit out Delayed Write Fails all over the place. I guess it just didn't like the server it was in or any other PCI slot in any other box for that matter. It doesn't even get past the initialization process before the system (whichever one it's in) POSTs.

Currently getting backups of the current server's disks on my system (typing on it right now) and going to move to a backup server. Good thing Progress is fairly stupid when it comes to all of this backup and restore business else I would be totally screwed.

I'm...not sure if I want to do IT any more after this. :p
 
Would also like to add that the "real" production server is now shut down and won't be coming back up for the foreseeable future. I thought it was going to play nice but it's pretty evident that it needs a complete nuke and pave along with some stress-testing before it comes back to a production environment.

IOW, Windows is fucked.

And to preempt it:

Whatsisname said:
WINDOZE BLOWZ LOONIX 4 LYFE M-FERZ
Click to expand...

:p
 
Damn - sorry to hear all that. Getting called in on a weekend blows. :( I worked for a hospital for quite a few years and it cam to the point that all servers were big box brands (IBM or Compaq) and we would only use the brands controllers, drives, etc. In the end it was a lot less problematic for us. They are not immune to failure by any means but when we had a high end IBM server, we knew with confidence that the IBM ServeRaid controller would play well in the machine, worked best in slot 4 (they tell you what slot to use), and if it failed IBM would have someone on site with a compatable replacement card in under 24 hours. Although since it was critical medical systems we actually kept our own complete systems as spares on site.

However - in my own shop here now, I have a hodgepodge of stuff running which at the moment is working well. ;) Good luck with the restores and rebuilds. A fresh loaded server is always nice.
 
I need to turn off disk write caching.

Doing it in Windows does nothing. It merely says that the setting cannot be changed because it's not allows. What the fuck does that mean?

Is there some registry setting that I can fanagle? This Delayed Write Caching bullshit has failed my databases twice now. I'd rather not have it happen a third time.
 
First: Adaptec? This is an LSI card, are you using LSI drivers? You might want to get the latest from LSI...

Second: Check the manual, specifically page 83 and section 5.10; I think write-through versus write-back may be the problem. The write-through setting should be the safer one. And if you haven't talked to LSI tech support, start *now*. You may also want to PM UICompE02; he knows his stuff when it comes to LSI.

Third: Jeez, I'm sorry this didn't come out as expected. :(

 
unhappy_mage said:
First: Adaptec? This is an LSI card, are you using LSI drivers? You might want to get the latest from LSI...

Second: Check the manual, specifically page 83 and section 5.10; I think write-through versus write-back may be the problem. The write-through setting should be the safer one. And if you haven't talked to LSI tech support, start *now*. You may also want to PM UICompE02; he knows his stuff when it comes to LSI.

Third: Jeez, I'm sorry this didn't come out as expected. :(

Click to expand...
Moved the DB to another server with an Adaptec 21something controller. Single channel...moved the disks from the production server over too, the ones that were in the test box got fucked somehow.

I don't really feel like talking about it right now. I was there for 18 hours and that's plenty for me.
 
You must log in or register to reply here.
Back
Top