<strong>My big question is...will apple release an xRAID that can be used for PowerMac TOWERS...not just for the xServe. Any speculation?</strong><hr></blockquote>
I don't think there's a reason you can't do that now. Just get a Fibrechannel card for your PM. I think. I could be wrong. It's been known to happen from time to time
<strong>What is indeed different and in favor of SCSI is the nominal speed of the interface. With SCSI topping at 160MB/s and ATA at 133/100. The main difference between the interfaces is that SCSI handles all the drives via the controller on the bus. Whereas ATA needs some processing power to manage the bus when there's more than one dive connected. (Master/Slave config) Therefor an ATA bus usually slows down when you connect a second drive and especially when copying from master to slave or vice versa. (It's similar to the ethernet hub vs. switch problem) On the other hand the SCSI Bus is vulnerable to malicious SCSI signal bursts generated by a defective controller. This usually slows down the chain, often bringing transfer to a halt or damaging data. ATA is less vulnerabel to this problems. There's also a simple splution to this when using ATA. Give each drive it's own bus. This gives you maximum security as a defective controller can't disturb other drives, maxes throughput to the mechanics possible maximum and doesn't eat processin power since there's no need for master/slave management on the bus. Due to much lower prices for ATA controllers this solution has many advantages over a SCSI chain.</strong><hr></blockquote>
So there's no CPU overhead when there's no slave drives? I thought the overhead was always there. Geez, I hope I can stop being wrong before my physics final tomorrow.
Oh, SCSI is hotswappable if your controller card and/or drivebay support(s) it, which most don't. I'm pretty sure about that. I'll ask my roommate, who's probably worked with such SCSI systems, when he gets back.
Many small hardware RAID setup use cheaper ATA drives and a SCSI converter. So internally ATA drives are used and externally a SCSI connection is made. Some PCI ATA controllers work that way as well.
Fiber Channel mainly is a means of transporting multiple protocol and data streams. Apple implements a so called double-speed Fiber channel allowing for 200MB/s. I don't know yet what protocol will be used for transfer. SCSI is likely but not guaranteed. No SCSI implementation can transfer 200MB/s and SCSI320 is only a per se standard. I haven't seen any real hardware. It's more likely that some kind of dual channel FireWire2 (800Mbit/s variant) or some fiber (optical) connection is used for that. But that's only an educated guess at the moment. Apple is very tight on prerelease info on th Xserve RAID.
Jeremiah Rich,
I don't guess you'll have to shell out 10k ($/?) for that baby. I guess 3k is more likely, but yet only a guesstimate.
TimTokyo,
Fibre Channel offers point-to-point, switched, and arbitrated loop interfaces. FC-Layers define the physical and protocol used for transmission.
FC-0 and FC-1 give details about the physical connection. (connectors, wires....)
FC-2 is the lower half of the Data Link layer. (Similar to the Media Acess Control in Ethernet)
FC-3 defines the services used on multiple Fiber channel ports. Like the use of dual channels to increase bandwidth.
FC-4 defines the high-layer protocol and it's mapping to to the transport layer.
there will only be one Xserve RAID. I'm sure you can connect that to your PowerMac if you expand it with an appropriate PCI card to match the interface of the RAID.
Whisper,
pay attention, that I did use the term "hot plug" opposed to "hot swap". Hot swapping means that you can cleanly detach a device form the active bus without causing electrical havoc. (It's a hardware only thing.) you can hot swap drives in a hardware RAID without the need to tell the OS that a drive is missing (depending on the RAID level of course) because the RAID processor handles this. Hot plugging means that the OS can also cleanly mount and unmount a device without the need to reboot. (eg. you can hot plug a FireWire HD/iPod)
More info on RAID Levels is in the works. Stay tuned!
Would it be possible to make the drives on the xRAID act as one big hardrive? ...Or have 8 of them act as one drive and the other 8 act as a mirror to the first 8? <img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" />
<strong>Would it be possible to make the drives on the xRAID act as one big hardrive? ...Or have 8 of them act as one drive and the other 8 act as a mirror to the first 8? <img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" /> </strong><hr></blockquote>
Um, both? That's what RAID means. (Except it'd be 7 and 7, there's only 14 drives.)
'Treating it as one big hard drive' would be RAID 0, where you are going to get a humongous throughput (14 separate ATA buses, it'd go at the max rate of the FiberChannel cables)
'Mirroring 7 drives to other seven drives' would be RAID 1.
<a href="http://www.3dscc.com/raid.htm" target="_blank">RAID Overview</a> It doesn't discuss Level 10, 30, or 50 though, where those are combinations '1 + 0', '3 + 0', and '5 + 0' if I understand correctly.
Yes that would be possible, see my RAID roundup below.
Nevyn,
Mirroring 7 drives to 7 other drives would not necessarily be L1 see below.
Raid Levels:
0: often called striping. Writes alternating chunks of data to alternating drives. Gives maximum speed, minumum safety. If a single drive of the RAID fails, all data is unrecoverable.
Read performance(total)=n x ReadPerformance(singleDrive)
Write performance same formula as read performance.
Minimum of 2 Drives: |A| |B|
1: Mirroring. Writes each block on all harddrives. Good speed in reading, slower speeds than a single drive when writing since it needs to write to each RAID member consequently. Good safety. Better safety with more drives.
Read performance(total) = n x ReadPerformance(singleDrive) (Since you have the same data on all drives, you can use all of them to read different chunks from, speeding reading up)
2,3,4: Mirroring with dedicated parity. Levels differ only in chunk size. L2: 1 bit, L3: 1 byte, L4: 4 bytes
Write performance even less than L1 since the parity has to be calculated and written as well.
Read performance slightly less because of parity as well.
Minimum set: 3 Drives: |A| |A| |pa|
5: Striping with interpersed parity.
minimum Set: 3 Drives: |A,pc| |B,pa| |C,pb|
You loose about one 8th of the total capacity to parity information for data safety.
6: Striping plus dedicated parity. Due to the extra drive with parity info a rebuild is faster but you loose one drive's capacity for safety and a little rebuild speed.
minimum set: |A,pc| |B,pa| |C,pb| |pa,pb,pc|
Levels 10, 20, 30... mean that there is a complete mirror of a whole RAID setup.
10: |
---------
| |
|A| |B| |A| |B|
The Xserve can do levels 0 and 1 in software. You can do this with any Mac OS X machine btw. Have a look at DiskUtility! Might come in for a surprise. :-) It's planned to support L10 in the Xserve as well since it would be perfectly possible with 4 drives. (Example above)
The Xserve RAID can do all RAID Levels in hardware and supports live rebuild.
Addition: Mac OS X can boot from any raid configuration and level.
Please feel free to add to and correct the charts above. I'm definitely too tired to proofread this now. |-)
Mirroring 7 drives to 7 other drives would not necessarily be L1 see below.
</strong><hr></blockquote>
I was thinking 7 independent drives mirrored to the seven other drives -> 7x RAID-1. That's what his quote meant to me. Shrug.
Any thoughts on what the theoretical max-throughput setup for this box would be? I mean, setting the whole XRaid up as one 14 disk RAID-0 (striping) setup would seem to me like it would max out the FiberChannel bus(es) given speedy disks....
[QB]Many small hardware RAID setup use cheaper ATA drives and a SCSI converter. So internally ATA drives are used and externally a SCSI connection is made. Some PCI ATA controllers work that way as well.
Yes, I believe that is the model that Apple will take with the xRaid.
Fiber Channel mainly is a means of transporting multiple protocol and data streams. SCSI is likely but not guaranteed. No SCSI implementation can transfer 200MB/s and SCSI320 is only a per se standard.
That is no problem for SCSI. SCSI will run just fine on the 2 GB fibre channel fabric.
Many vendors, EMC, IBM, HP, etc have products in production right now that do this. This is because
with SCSI on FibreChannel, you must distinguish
between the physical layer and protocol layer.
SCSI on FibreChannel will run as fast as the
fibrechannel implementation (the physical layer for SCSI in this case).
No there is no ATA protocol layer for FC. That makes it very likely that Apple will use SCSI as the encapsulation. In that case the XRaid's ATA-disks will be presented to OSX as scsi devices.
Would Apple be adverse to using the SCSI protocol?
Comments
<strong>My big question is...will apple release an xRAID that can be used for PowerMac TOWERS...not just for the xServe. Any speculation?</strong><hr></blockquote>
I don't think there's a reason you can't do that now. Just get a Fibrechannel card for your PM. I think. I could be wrong. It's been known to happen from time to time
<strong>What is indeed different and in favor of SCSI is the nominal speed of the interface. With SCSI topping at 160MB/s and ATA at 133/100. The main difference between the interfaces is that SCSI handles all the drives via the controller on the bus. Whereas ATA needs some processing power to manage the bus when there's more than one dive connected. (Master/Slave config) Therefor an ATA bus usually slows down when you connect a second drive and especially when copying from master to slave or vice versa. (It's similar to the ethernet hub vs. switch problem) On the other hand the SCSI Bus is vulnerable to malicious SCSI signal bursts generated by a defective controller. This usually slows down the chain, often bringing transfer to a halt or damaging data. ATA is less vulnerabel to this problems. There's also a simple splution to this when using ATA. Give each drive it's own bus. This gives you maximum security as a defective controller can't disturb other drives, maxes throughput to the mechanics possible maximum and doesn't eat processin power since there's no need for master/slave management on the bus. Due to much lower prices for ATA controllers this solution has many advantages over a SCSI chain.</strong><hr></blockquote>
So there's no CPU overhead when there's no slave drives? I thought the overhead was always there. Geez, I hope I can stop being wrong before my physics final tomorrow.
Oh, SCSI is hotswappable if your controller card and/or drivebay support(s) it, which most don't. I'm pretty sure about that. I'll ask my roommate, who's probably worked with such SCSI systems, when he gets back.
Fiber Channel mainly is a means of transporting multiple protocol and data streams. Apple implements a so called double-speed Fiber channel allowing for 200MB/s. I don't know yet what protocol will be used for transfer. SCSI is likely but not guaranteed. No SCSI implementation can transfer 200MB/s and SCSI320 is only a per se standard. I haven't seen any real hardware. It's more likely that some kind of dual channel FireWire2 (800Mbit/s variant) or some fiber (optical) connection is used for that. But that's only an educated guess at the moment. Apple is very tight on prerelease info on th Xserve RAID.
Jeremiah Rich,
I don't guess you'll have to shell out 10k ($/?) for that baby. I guess 3k is more likely, but yet only a guesstimate.
TimTokyo,
Fibre Channel offers point-to-point, switched, and arbitrated loop interfaces. FC-Layers define the physical and protocol used for transmission.
FC-0 and FC-1 give details about the physical connection. (connectors, wires....)
FC-2 is the lower half of the Data Link layer. (Similar to the Media Acess Control in Ethernet)
FC-3 defines the services used on multiple Fiber channel ports. Like the use of dual channels to increase bandwidth.
FC-4 defines the high-layer protocol and it's mapping to to the transport layer.
I know of SCSI, HIPPI and FDDI protocol layers. I'm not aware of an ATA protocol layer. But who knows what Apple wil use for it. SCSI is unlikely <a href="http://www.geocities.com/SouthBeach/Lagoon/9819/acronyms.html" target="_blank">IMHO</a>.
duplicity,
there will only be one Xserve RAID. I'm sure you can connect that to your PowerMac if you expand it with an appropriate PCI card to match the interface of the RAID.
Whisper,
pay attention, that I did use the term "hot plug" opposed to "hot swap". Hot swapping means that you can cleanly detach a device form the active bus without causing electrical havoc. (It's a hardware only thing.) you can hot swap drives in a hardware RAID without the need to tell the OS that a drive is missing (depending on the RAID level of course) because the RAID processor handles this. Hot plugging means that the OS can also cleanly mount and unmount a device without the need to reboot. (eg. you can hot plug a FireWire HD/iPod)
More info on RAID Levels is in the works. Stay tuned!
<strong>Would it be possible to make the drives on the xRAID act as one big hardrive? ...Or have 8 of them act as one drive and the other 8 act as a mirror to the first 8? <img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" /> </strong><hr></blockquote>
Um, both? That's what RAID means. (Except it'd be 7 and 7, there's only 14 drives.)
'Treating it as one big hard drive' would be RAID 0, where you are going to get a humongous throughput (14 separate ATA buses, it'd go at the max rate of the FiberChannel cables)
'Mirroring 7 drives to other seven drives' would be RAID 1.
<a href="http://www.3dscc.com/raid.htm" target="_blank">RAID Overview</a> It doesn't discuss Level 10, 30, or 50 though, where those are combinations '1 + 0', '3 + 0', and '5 + 0' if I understand correctly.
Yes that would be possible, see my RAID roundup below.
Nevyn,
Mirroring 7 drives to 7 other drives would not necessarily be L1 see below.
Raid Levels:
0: often called striping. Writes alternating chunks of data to alternating drives. Gives maximum speed, minumum safety. If a single drive of the RAID fails, all data is unrecoverable.
Read performance(total)=n x ReadPerformance(singleDrive)
Write performance same formula as read performance.
Minimum of 2 Drives: |A| |B|
1: Mirroring. Writes each block on all harddrives. Good speed in reading, slower speeds than a single drive when writing since it needs to write to each RAID member consequently. Good safety. Better safety with more drives.
Read performance(total) = n x ReadPerformance(singleDrive) (Since you have the same data on all drives, you can use all of them to read different chunks from, speeding reading up)
Write performance= WritePerformance(SingleDrive)/n
(All drives have to be written to)
Minimum set: 2 Drives: |A| |A|
2,3,4: Mirroring with dedicated parity. Levels differ only in chunk size. L2: 1 bit, L3: 1 byte, L4: 4 bytes
Write performance even less than L1 since the parity has to be calculated and written as well.
Read performance slightly less because of parity as well.
Minimum set: 3 Drives: |A| |A| |pa|
5: Striping with interpersed parity.
minimum Set: 3 Drives: |A,pc| |B,pa| |C,pb|
You loose about one 8th of the total capacity to parity information for data safety.
6: Striping plus dedicated parity. Due to the extra drive with parity info a rebuild is faster but you loose one drive's capacity for safety and a little rebuild speed.
minimum set: |A,pc| |B,pa| |C,pb| |pa,pb,pc|
Levels 10, 20, 30... mean that there is a complete mirror of a whole RAID setup.
10: |
---------
| |
|A| |B| |A| |B|
The Xserve can do levels 0 and 1 in software. You can do this with any Mac OS X machine btw. Have a look at DiskUtility! Might come in for a surprise. :-) It's planned to support L10 in the Xserve as well since it would be perfectly possible with 4 drives. (Example above)
The Xserve RAID can do all RAID Levels in hardware and supports live rebuild.
Addition: Mac OS X can boot from any raid configuration and level.
Please feel free to add to and correct the charts above. I'm definitely too tired to proofread this now. |-)
<strong>Nevyn,
Mirroring 7 drives to 7 other drives would not necessarily be L1 see below.
</strong><hr></blockquote>
I was thinking 7 independent drives mirrored to the seven other drives -> 7x RAID-1. That's what his quote meant to me. Shrug.
Any thoughts on what the theoretical max-throughput setup for this box would be? I mean, setting the whole XRaid up as one 14 disk RAID-0 (striping) setup would seem to me like it would max out the FiberChannel bus(es) given speedy disks....
[QB]Many small hardware RAID setup use cheaper ATA drives and a SCSI converter. So internally ATA drives are used and externally a SCSI connection is made. Some PCI ATA controllers work that way as well.
Yes, I believe that is the model that Apple will take with the xRaid.
Fiber Channel mainly is a means of transporting multiple protocol and data streams. SCSI is likely but not guaranteed. No SCSI implementation can transfer 200MB/s and SCSI320 is only a per se standard.
That is no problem for SCSI. SCSI will run just fine on the 2 GB fibre channel fabric.
Many vendors, EMC, IBM, HP, etc have products in production right now that do this. This is because
with SCSI on FibreChannel, you must distinguish
between the physical layer and protocol layer.
SCSI on FibreChannel will run as fast as the
fibrechannel implementation (the physical layer for SCSI in this case).
I know of SCSI, HIPPI and FDDI protocol layers. I'm not aware of an ATA protocol layer. But who knows what Apple wil use for it. SCSI is unlikely <a href="http://www.geocities.com/SouthBeach/Lagoon/9819/acronyms.html" target="_blank">IMHO</a>.
No there is no ATA protocol layer for FC. That makes it very likely that Apple will use SCSI as the encapsulation. In that case the XRaid's ATA-disks will be presented to OSX as scsi devices.
Would Apple be adverse to using the SCSI protocol?