Setting up a RAID 5 array w/ 4-channel SATA card

E

Elledan

[H]ard|DCer of the Month - April 2010
Joined
Oct 18, 2001
Messages
15,913
First of all, I don't have any experience with RAID 5 whatsoever, so I might say some seriously stupid things :)

Anyway, assuming that the smallest RAID 5 array requires three disks (all the same size?), a controller card with at least three channels is required. Of course, this means that one gets a four channel card (I've been looking at the 3Ware 8506-4LP), leaving one channel free.

So, my question is basically whether this channel is somehow 'wasted', or that it can still be used (as part of the array?).

Also, with multiple controller cards in one system, is each card good for a single RAID 5 array, or can they be combined?

Finally, has anyone here had trouble with a SATA RAID setup? Especially in regards to the physical parts, i.e. cables (tabs breaking off, connectors getting loose, etc.).

Thanks in advance for tolerating those questions!
 
i don't know because i myself haven't done anything raid 5 yet... but from what i understand 3 drives is only the minimum you should be able to use all 4 channels in your raid 5 setup it would just write like this part1-drive1 part2-drive2 part3-drive3 part11-drive4 part22-drive1 part33-drive2

get it? 3 is only the MINIMUM number of drives needed it doesn't mean that you have to build it 3 drives at a time...

parity just requires 3 drive minimum so that you can always figure out what the full picture is no matter what drive goes bad... the principal is this

1 + x = 5 so x = 4

or 2 + 7 = x so x = 9

or x + 8 = 11 so x = 3

understand how it works now?

i didn't know it but there are cards that can do raid 5 with mirroring also, i just noticed this the other week... but that of course means a MINIMUM of 6 drives... HAHA!!!
 
wow, im setting up a raid 5 setup as we speak....I'm doing it for the redundancy and speed. As far as i know, there is very little wasted space...get a card with a xor processor, or you will be hurting..like the promise fastrak on newegg (168.00)
 
LeadMetal1402 said:
wow, im setting up a raid 5 setup as we speak....I'm doing it for the redundancy and speed. As far as i know, there is very little wasted space...get a card with a xor processor, or you will be hurting..like the promise fastrak on newegg (168.00)
Click to expand...
Well, it appears that the success rate with 3Ware cards is far higher than with Promise and others (e.g. today on /. ).

I'd far rather spend a bit more for a properly working setup, than suffer the results of going for the 'cheap' solution.

Anyway, RAID 5 with a 4-channel card is probably best set up as 3 data/parity-drives and 1 hot-spare. Assuming 160 GB HDDs, this will provide (3 - 1) * 160 = 320 GB of space.

BTW, RAID 5+1 would be kind of nice. I wonder if two RAID 5 arrays can be duplexed into a RAID 1 array? ;)
 
actually i spoke in error... i just realized while re-reading this post... 0 = striping and 1 = redundancy so the cards that support raid50 is actually striping? now i'm kind of confused... because i thought parity was in a way striping but with fault tolerance... so what would the benefit of raid50 be then?
 
absolut][krypkn0t said:
actually i spoke in error... i just realized while re-reading this post... 0 = striping and 1 = redundancy so the cards that support raid50 is actually striping?
Click to expand...
Yes, RAID 50 means RAID 5 (distributed parity) of which stripes are distributed like with RAID 0, increasing write and read speeds.
 
as Elleden pointed out ;)
RAID 5+0 & 0+5 and there is 51 or 15 as well

Id suggest that at least the home user to small business having a hotspare is a waste
as the need to have the data "online" isnt as extreme, its generally acceptable to "discover" the single failure and replace the drive, then rebuild
you wieght that immediate rebuild window against the extra storage efficiency offered by populating an additional channel with a drive (Capacity being number of drives minus one)

regarding the "cheap" IDE drives, yup 3wares are nice, but not all the Promise line for instance, lack a dedicated XOR processor, and sport quite large caches, Ive had no reason to regrets* getting my SX6000 and hear the same about the SX4000
(*with the noteable exception on a small lie in the manual later fixed with a BIOS upgrade but I adopted early in the line) but then niether are "cheap" :p

its possible to "plaid" drives, which Ive seen mentioned in enterprise storage circles
that being employing software and hadware RAID, for instance 2 RAID 5 cards and both of those drives as Dyanamic Disk RAID1 volumes

you can see the complexity issues Im sure
I know of no cards that duplex\combine across the PCI bus, but check with UICompE02 or Snugglebear for a definitive answer on that
I do have a dual RAID card subsystem, which can be employed as either dual channel (and do RAID 50) or as a fail over, but thats a SCSI subsystem external
 
Ice Czar said:
as Elleden pointed out ;)
Click to expand...
Hey, I'm not calling you Ice Czer, am I? :( ;)

Id suggest that at least the home user to small business having a hotspare is a waste
as the need to have the data "online" isnt as extreme, its generally acceptable to "discover" the single failure and replace the drive, then rebuild
you wieght that immediate rebuild window against the extra storage efficiency offered by populating an additional channel with a drive (Capacity being number of drives minus one)
Click to expand...
Good point. Using the example with the 160 GB HDDs again, that would mean (4 - 1) * 160 = 480 GB. Much more efficient than setting up a second array, not to mention the increase in read speed.

BTW, provided that all (SATA) HDDs are installed in hot-swap cradles, does the failure of a single HDD in a RAID 5 array mean that the array will still be online while the dead drive is being replaced?

regarding the "cheap" IDE drives, yup 3wares are nice, but not all the Promise line for instance, lack a dedicated XOR processor, and sport quite large caches, Ive had no reason to regrets* getting my SX6000 and hear the same about the SX4000
(*with the noteable exception on a small lie in the manual later fixed with a BIOS upgrade but I adopted early in the line) but then niether are "cheap" :p
Click to expand...
Taking a look at the Promise FastTrak S150 SX4, it certainly looks quite good. It definitely has all the features I would need. Only thing I don't know is how good the Linux drivers are for this card.

Buggy drivers are really the last thing I need, which is why I was looking at 3Ware cards.
 
Elledan said:
BTW, provided that all (SATA) HDDs are installed in hot-swap cradles, does the failure of a single HDD in a RAID 5 array mean that the array will still be online while the dead drive is being replaced?
Click to expand...

it would be online, but with no redundancy, loose another and the game is over ;)

sorry about the typo, Im normally reread and edit liberally but was in a hurry this morning
and I average at least a few typo per paragraph :p

The Promise SATA line doent seem to have the dedicated RISC Processors
instead some sort of IC XOR engine and while there probably isnt a whiole lot of difference I think they do impact the CPU more? Hard for me to judge, and Ive seen few reviews, but the one I do know is > http://www.tweakers.net/reviews/400

in Danish ;) and Femme Taken built his suite of Benchmarks out of IPEAK SPT just like Storage Reviews testbed ;)

benchmarks start in earnest on page 7
translation by url with worldlingo is pretty good
 
Ice Czar said:
it would be online, but with no redundancy, loose another and the game is over
Click to expand...
Cool. It's a lot better than having no redundancy to begin with.


(BTW, it's 'lose'. Sorry, it's a pet peeve =) )

sorry about the typo, Im normally reread and edit liberally but was in a hurry this morning
and I average at least a few typo per paragraph :p
Click to expand...
It's alright :)

The Promise SATA line doent seem to have the dedicated RISC Processors
instead some sort of IC XOR engine and while there probably isnt a whiole lot of difference I think they do impact the CPU more? Hard for me to judge, and Ive seen few reviews, but the one I do know is > http://www.tweakers.net/reviews/400

in Danish ;) and Femme Taken built his suite of Benchmarks out of IPEAK SPT just like Storage Reviews testbed
Click to expand...
So, that would be the big difference between Promise and 3Ware cards. Promise is apparently still aiming for the lower-end server and workstation market. I like the high-end, though, and I'm not afraid to spend some more money.

Ah, and the review is in Dutch, not Danish. Coincidentally, it's one of the languages I speak, so no need for me to translate it =)
 
I just read the article you linked to, Ice Czar.

It pretty much confirmed my thoughts regarding Promise: their products are unsuitable for any professional applications.

The 3ware card (this was an older one in the 8xxx series, though) showed that having 2 MB (!) cache memory on a RAID card is more than just a bad idea, as well that it lacks many features found on the cheaper cards. It seems that 3ware uses the 8xxx series merely as a way to get more people to use their 9xxx series of cards.

All in all, I have to say that the LSI MegaRAID 150-4 card looks like the most attractive option. It performs very good in RAID 5 (better than the Promise S150 SX4) and has proper driver support for a number of platforms, including Linux.

Now I just have to find where I can buy this card :)
 
Ice Czar said:
what is the application by the way?
Click to expand...
Fileserver for a small LAN. Basically I've got one workstation, a couple of Linux systems and an HTPC (well, at least I like to pretend that it is one :p ) for which I wish to centralize all data, for the sake of preventing data-loss without inflicting serious damage to my wallet (i.e. having RAID 1/5 arrays in every system).

The fileserver will store source code and documentation of projects I'm working on, video- and audio-files, backups and various other files.
 
In that case, and if you can afford it Id agree with you the 3ware looks good

had another option but its strickly for large file video capture\editing type transfers
doesnt work a damn with small incremental "data base" type transfers
 
Ice Czar said:
In that case, and if you can afford it Id agree with you the 3ware looks good
Click to expand...
That is, assuming that it has more than 2 MB cache memory.

Regardless, LSI is another option, if only because it seems to perform at least as good, if not better than the 3ware 8xxx series, while it's a lot cheaper.

had another option but its strickly for large file video capture\editing type transfers
doesnt work a damn with small incremental "data base" type transfers
Click to expand...
Well, tell us anyway. Might be useful for another project :)
 
As an eBay Associate, HardForum may earn from qualifying purchases.
Elledan said:
Well, tell us anyway. Might be useful for another project :)
Click to expand...

Oh its that Netcell SynchRAID, an optimized RAID3 in effect
http://www.netcell.com/technology.htm
download the Whitepaper at the bottom
"SynchRAID Removing RAID Adapter Bottlenecks" (PDF)
they come in a 3 channel and 5 channel in PATA or SATA
but at 32bit 33MHz only
specifically aimed at workstation and desktop, for CAD and Video\Image Production

Data Paths and Word Level vs Block Striping
SyncRAID supports a wide data path to and from the drive array, which is an integer sum of the individual widths of the individual drive I/O interfaces. This allows much Higher data throughput, especially when streaming large data files on and off the disk as in emerging 64-bit CPU desktop and workstation environments.

Figure 1 illustrates how 64-bit data word destined for the drive array in a SynchRAID controller is handled versus in a legacy controller. For a legacy RAID 16-bit ATA implementation, a block of 64-bit words from the host bus is divided into a stream of 16-bit words and issued to individual drives until the block stripe is full or the block write complete. If the block occupies more than a single drive stripe block or "chunk" then it moves onto the next drive to start using another stripe block until the block transfer has completed. As we shall see later, the parity generation or update method very much dictates the performance of array write operations.

SynchRAID, due to its innovative synchronization scheme, removes this data and blocking arrangement altogether, keeping the data words intact as 64 bits all the way to the drives. This effectively creates a 4x-speed ATA interface.*(with 4 drives) Applying the concept to multiple SATA drives, the interface becomes a 4x SATA interface, and so on.
Click to expand...

of course if the data transfer falls below 64bit.... :p
you see why its sort of "special"

that of course doesnt address evn half of it, but is a taste
effectively its supposed to be RAID 0 performance with redundancy
Tom's reviewed it, but then it was Tom's that reviewed it :p
 
Elledan said:
Thank you for the offer. Unfortunately I don't have the funds for this project yet (another two months or so) :(
Click to expand...

Well, despite the high number of people watching the auction (yeah ebay lets you see that, kinda nifty), there were no takers. So I'll probably relist the card and the drives separately in a week or so to see if there's a more positive response.

Then again, maybe i should just keep this gear and build another box just for the heck of it. ;)
 
infocalypse said:
Then again, maybe i should just keep this gear and build another box just for the heck of it. ;)
Click to expand...
More HDD space is always a good thing, isn't it? :)
 
Ice Czar said:
of course if the data transfer falls below 64bit.... :p
you see why its sort of "special"
Click to expand...
Indeed. It appears to be aimed at a very specific market. It's definitely not as general purpose as 99% of the RAID controllers out there.

that of course doesnt address evn half of it, but is a taste
effectively its supposed to be RAID 0 performance with redundancy
Click to expand...
Something tells me that by the time I might actually have a need for this product, we'll all be using M/NRAM-based solid state disks ;)
 
You must log in or register to reply here.
Back
Top