3 Hard Drives?

[JCLW]

RAID 1 for security. Speed doesn't really matter to me - currently my RAID 1 array is running off of 5400rpm drives.

What you were talking about previously (2x 80bg stripped + 1x 160gb) is similar to RAID 10 (except with three drives). The thing is that it's not really true RAID security - if you lost a 80gb drive between "backups" then you'd lose everything since your last backup. If I work at my computer all day long then it only gets a chance to "backup" every night - I could lose a whole days work ($40/hr x 8hrs = $320). Also the added stress of doing an array mirror would put a lot of extra wear and tear on the drives.

Eventually I'll move to a larger array (probably six drive RAID 5) in a seperate file server to try and cut down on the noise and heat from my workstation.

Also...
The three drive stripped theory sounds good, but has a flaw. It's the same reason going RAID 1 (mirror) doesn't improve read speeds. Imagine you have two identical drives, capable of reading one bit in one ms (not realistic, but it'll do for the example) as the platter moves past the head. This means the drive can read 1000 sequential bits in one second (I warned you it wasn't realistic). But if we want to attempt to use it as a stripped array for reading we only want to read every other bit (1,3,5,7...999 on drive 1 and 2,4,6,8...1000 on drive two). But it would still take one second for each drive to get from 1 to 1000, regardless of how many bits the head read, because the head can't jump from 1 to 3 - it has to wait for two to go past first. It's just a physical limitation of the drive that can only be overcome by increasing the data density and/or speeding up the platter rotation to the point where the drive could sustain a higher data transfer rate then the interface could handle (150mb/s for SATA).

I didn't do the best job explaining that - Hope I didn't just confuse everyone.

 

[JCLW]

Here's a pic of my workstation (because everyone likes pictures).

Dual PIII-800E on a P2B-D (i440BX) board.

Drives top to bottom:
Plextor 708A DVD +/-RW - Onboard primary IDE (Master)
Pioneer DVD-500M - Onboard secondary IDE (Master)
WD 80GB 7200RPM 8MB - Promise TX2000 primary IDE (Master)
Maxtor 40GB 5400RPM 2MB - Promise TX2000 secondary IDE (Master)
Maxtor 40GB 5400RPM 2MB - Promise TX2000 secondary IDE (Slave)

The two 40GBs make up my RAID 1 data array. You can see the Promise TX2000 below the video card.

 

[JCLW]

For me RAID1 is a waist of money.

It depends what you store I guess. I make a living off of this computer and depending on what kind of a project I'm working on the data on that RAID 1 array can be worth over $5000 to me. So ~$100 for the RAID card and ~$150 for the 2nd drive (at the time) isn't much to pay for the security. I'll also do offsite backups once a week in a big project.

The reason I went to a RAID 1 array was because I once lost a drive with all my data on it.

 

[Sjaak]

[JCLW theory]

[JCLW pic]

I dont agree on the drive part...IMO, the controller splits the file into two parts (1000 bit file, 500/500) then it has every drive write one part from the beginning of the drive (drive 1 = bit 1 to 500, drive 2 = bit 501 to 1000 etc) Thereby, the speed would be (theoretically) doubled, because both drives operate independently of the data.

Nice pic...looks kinda dusty though *cough, cough*

 

[JigPu]

The three drive stripped theory sounds good, but has a flaw. It's the same reason going RAID 1 (mirror) doesn't improve read speeds. Imagine you have two identical drives, capable of reading one bit in one ms (not realistic, but it'll do for the example) as the platter moves past the head. This means the drive can read 1000 sequential bits in one second (I warned you it wasn't realistic). But if we want to attempt to use it as a stripped array for reading we only want to read every other bit (1,3,5,7...999 on drive 1 and 2,4,6,8...1000 on drive two). But it would still take one second for each drive to get from 1 to 1000, regardless of how many bits the head read, because the head can't jump from 1 to 3 - it has to wait for two to go past first. It's just a physical limitation of the drive that can only be overcome by increasing the data density and/or speeding up the platter rotation to the point where the drive could sustain a higher data transfer rate then the interface could handle (150mb/s for SATA).

I didn't do the best job explaining that - Hope I didn't just confuse everyone.

Nope, that finally explains to me just why RAID 1 dosen't have awesome read speeds

Though if the RAID array could be set up right, you could actually have performances increases as fragmentation increases by having one drive seek to the second cluster of a file while the other drive reads the first cluster. It wouldn't help at all in sequential reads (probably would hurt it actually), though not having a drive slow down in fragmented conditions would be rather cool

JigPu

 

[Sjaak]

[...]

[fictional conversation]

<- Dude whatsup with your pc? its so slow

-> Ye, my drive is getting sloppy, i gotta run the windows fragmentation tool more often

[/fictional conversation]

 

[forbin rhodes]

I have a ic7max3 board with silicon image four disc raid. Presently running 2 36 gb
raptors in raid 0. Burst speed 109 mbs read speed 101 mbs. Have third raptor on its way that should push the speed close to the max of 150 mbs.

 

[JCLW]

I dont agree on the drive part...IMO, the controller splits the file into two parts (1000 bit file, 500/500) then it has every drive write one part from the beginning of the drive (drive 1 = bit 1 to 500, drive 2 = bit 501 to 1000 etc) Thereby, the speed would be (theoretically) doubled, because both drives operate independently of the data.

Nice pic...looks kinda dusty though *cough, cough*

When you set up an array you have to pick the stripe size - 16k is typical. So bits 0k-15k would be on drive 1, bits 16k-31k would be on drive 2, bits 32k-47k would be on drive 1, etc. The controller has no idea what size the file is, all it does is write 16k to one drive, and then 16k to the other.

It turns five years old in the fall - I'll clean it then.

Though if the RAID array could be set up right, you could actually have performances increases as fragmentation increases by having one drive seek to the second cluster of a file while the other drive reads the first cluster. It wouldn't help at all in sequential reads (probably would hurt it actually), though not having a drive slow down in fragmented conditions would be rather cool

JigPu

Yes, but you'd need a fairly sophisticated RAID controller. It would have to keep an up to date cached copy of the drive's file allocation table at all times to know where to position the second drive's head as soon as the first drive started reading the first part of the file. If you could do that you could probably even do what Sjaak was suggesting with non-fragmented files, and have the second drive start reading halfway through the length of the file so that by the time the first drive got halfway through the file the 2nd half was already read and cached by the 2nd drive.

It would be nice, but I don't think it's possible with today's software and hardware.

-----

On a completely different note there is talk of drive manufacturers looking at doing "internal stripping" in their drives. Basically they currently write to one side of the first platter, then go on to the other side, then on to the 2nd platter... It would be pretty easy to alternate writting between the two sides of each platter, so you end up with a RAID 0 (stripped) array inside the disk itself, which would be transparent to the controller or the OS. It might even be possible to stripe not only between sides of the same platter but between platters themselves in multi-platter drives, effectively giving you the equivalent of a four or six drive RAID 0 (striped) array with a one bit stripe size . It would also be just as safe as a single disk. If current drives are offering 40mb/s a dual platter internally stripped 7200rpm drive could theoretically offer something along the lines of a 160mb/s sustained transfer rate.

 

[JCLW]

I have a ic7max3 board with silicon image four disc raid. Presently running 2 36 gb
raptors in raid 0. Burst speed 109 mbs read speed 101 mbs. Have third raptor on its way that should push the speed close to the max of 150 mbs.

The 150mb/s is a value for each drive, so you won't have to worry about that. Your limiting factor will probably be the PCI bus that the Sil controller is hanging off of. Max PCI bandwidth is about 133mb/s, theoretical is more like 125mb/s. Once we get some PCI-E controllers that'll get bumped up to 250mb/s.

Still, that'll be a crazy fast drive setup you have there

 

[Sjaak]

Hell yes.

Too bad raptors still suck at $/Gbyte

 

[forbin rhodes]

109 dollars for a 36 gb raptor is cheap. I paid 10,000 dollars for a winchester 5 mb
drive in 1979. It was definitly bigger than a breadbox.

 

[Sjaak]

Seeing that i can get a hitachi deskstar 160GB sata for 99€, which is more quiet and in some benchmarks faster then the raptor, there is no point in buying. The raptors here do 110€ / 200€ for the 37/74GB versions