Coss Posted October 2, 2003 Author Report Share Posted October 2, 2003 OSX 10.2.8 2 x 120GB Seagate 8MB buffer each with a caloptic fire wire bridge, each with a single firewire cable into a 6 port firewire repeater hub and thence one cable to the single firewire port (400) on my TiBook. The disks are striped with Apple's raid. But wait, I've just got 2 120GB Western Digitals and 2 more firewire bridges. I'm gonna test the 4 disks in various combinations with Apples raid and see what's what. Results soon. Cheers Coss Link to comment Share on other sites More sharing options...
think_too_mut Posted October 2, 2003 Report Share Posted October 2, 2003 >I would not even consider it unless I wanted a robust RAID 5 solution: in that case, a major gain is possible because the mirroring can be invisible to the PC. RAID 5 is striping with parity distributed accross multiple drives. No mirroring. RAID 1 is mirroring. RAID 10 is striping + mirroring. There is no robust RAID 5 solution unless parity is calculated on the array. Anything done through hosts's CPU is Mickey Mouse. Link to comment Share on other sites More sharing options...
BritTim Posted October 3, 2003 Report Share Posted October 3, 2003 RAID 5 is striping with parity distributed accross multiple drives. No mirroring. I am sorry. You are right and I knew that. When I wrote that post I was suffering from lack of sleep. My point was supposed to be that the overhead involved in providing for data integrity can be offloaded from the CPU (yes, it is done with parity bytes, not disk mirroring). As you say, for ultimate reliability reasons (as well as performance) RAID 5 should be implemented by the disk subsystem rather than software in the host CPU. Link to comment Share on other sites More sharing options...
think_too_mut Posted October 3, 2003 Report Share Posted October 3, 2003 >RAID 5 should be implemented by the disk subsystem rather than software in the host CPU. Not directed to you, for other readers' information - with striping, one "write" operation consists of 7 i/o operations. Read has 4 operations. A massive i/o as with video editing/streaming is a huge burden to host's CPU even if it is the only thing happening at a time. For single user systems, as Mac, it will work more or less reliably. Anything more than that would require additional processing power, ideally within a disk array itself. Link to comment Share on other sites More sharing options...
Coss Posted October 3, 2003 Author Report Share Posted October 3, 2003 2 x Western Digital 120GB 2 MB cache striped as 240GB Disk Test 62.22 Sequential 46.29 Uncached Write 69.51 28.98 MB/sec [4K blocks] Uncached Write 65.68 26.90 MB/sec [256K blocks] Uncached Read 27.01 4.28 MB/sec [4K blocks] Uncached Read 50.56 20.43 MB/sec [256K blocks] Random 94.85 Uncached Write 193.00 2.90 MB/sec [4K blocks] Uncached Write 112.69 25.41 MB/sec [256K blocks] Uncached Read 81.44 0.54 MB/sec [4K blocks] Uncached Read 63.13 12.99 MB/sec [256K blocks] 2 x Seagate 120GB 8 MB cache striped as 240GB Disk Test 63.15 Sequential 46.18 Uncached Write 69.58 29.00 MB/sec [4K blocks] Uncached Write 65.53 26.83 MB/sec [256K blocks] Uncached Read 26.43 4.18 MB/sec [4K blocks] Uncached Read 52.24 21.11 MB/sec [256K blocks] Random 99.81 Uncached Write 171.93 2.58 MB/sec [4K blocks] Uncached Write 112.58 25.39 MB/sec [256K blocks] Uncached Read 96.05 0.63 MB/sec [4K blocks] Uncached Read 66.81 13.75 MB/sec [256K blocks] 4 x 120GB striped as 480GB Disk Test 63.57 Sequential 44.15 Uncached Write 55.94 23.32 MB/sec [4K blocks] Uncached Write 66.70 27.31 MB/sec [256K blocks] Uncached Read 26.34 4.17 MB/sec [4K blocks] Uncached Read 50.59 20.44 MB/sec [256K blocks] Random 113.47 Uncached Write 329.81 4.95 MB/sec [4K blocks] Uncached Write 115.68 26.09 MB/sec [256K blocks] Uncached Read 94.82 0.63 MB/sec [4K blocks] Uncached Read 76.75 15.80 MB/sec [256K blocks] 4 x 120GB mirrored as 120GB Disk Test 32.46 Sequential 22.55 Uncached Write 18.88 7.87 MB/sec [4K blocks] Uncached Write 18.57 7.60 MB/sec [256K blocks] Uncached Read 22.19 3.51 MB/sec [4K blocks] Uncached Read 39.19 15.84 MB/sec [256K blocks] Random 57.89 Uncached Write 80.78 1.21 MB/sec [4K blocks] Uncached Write 34.02 7.67 MB/sec [256K blocks] Uncached Read 81.98 0.54 MB/sec [4K blocks] Uncached Read 66.14 13.61 MB/sec [256K blocks] I've just found out that if I were to get another Firewire bus, i.e. a PCMCIA CardBus, and striped the disks across that and the internal bus, then I'd get the results I'm expecting. I think I was told that earlier in the thread, but I'm slow on the uptake... Cheers Coss Link to comment Share on other sites More sharing options...
Guest Posted October 4, 2003 Report Share Posted October 4, 2003 1394 controllers can be bus masters just as ATA controllers can, so differences in disk performance would likely be due to other factors. Indeed, a quick test with copying a few massive files between a 1394 drive and a Serial ATA drive shows sustained transfer at about 3% CPU usage. This is on a P4 3GHz, Asus P4C800 motherboard with integrated 1394 and Serial ATA, under Windows XP Pro. Link to comment Share on other sites More sharing options...
Guest Posted October 4, 2003 Report Share Posted October 4, 2003 Yeah! I would be unlikely to use RAID5 where performance was the critical factor. At the 10000-foot level, each write operation involves reading all of the other corresponding pieces of the stripe, then calculating new checksum/parity data, then writing *both* the data that was the original subject of the i/o operation *and* the new checksum/parity data. Offloading this to external hardware certainly helps, but there's no getting around that RAID5 is simply not the right tool for that kind of job. On the other hand, a stripe set *without* parity has the potential to boost read and write performance, subject to all sorts of other variables that people have mentioned in this thread. Link to comment Share on other sites More sharing options...
Coss Posted October 8, 2003 Author Report Share Posted October 8, 2003 And the results are in. I wont reproduce the entire results here, just the scores from XBench, and I've got a 50% increase in speed overall. All of these on a Tibook 667MHz OSX 10.2.8 Test one: 2 x 120GB WDigital 2MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing only the inbuilt firewire bus. Score = 60.25 2 x 120GB Seagate 8MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing only the inbuilt firewire bus. Score = 61.47 At this point I installed the 10.2.8 'fix-up' that Apple released after the first 10.2.8 was withdrawn after I'd installed it. Test two: 2 x 120GB WDigital 2MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing only the inbuilt firewire bus, but with a second bus (PCMCIA cardbus firewire) plugged in and empty. Score = 78.49 2 x 120GB Seagate 8MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing only the inbuilt firewire bus, but with a second bus (PCMCIA cardbus firewire) plugged in and empty. Score = 79.68 Test three: 2 x 120GB WDigital 2MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing the inbuilt firewire bus and a second bus (PCMCIA cardbus firewire). Score = 91.7 2 x 120GB Seagate 8MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing only the inbuilt firewire buss, but with a second bus (PCMCIA cardbus firewire) plugged in and empty. Score = 81.82 Test four: 2 x 120GB WDigital 2MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing the inbuilt firewire bus and a second bus (PCMCIA cardbus firewire). Score = 91.51 2 x 120GB Seagate 8MB cache striped with Apple's RAID through a firewire hub/repeater, utilizing the inbuilt firewire bus and a second bus (PCMCIA cardbus firewire). Score = 92.21 It would seem that a combination of the 10.2.8 'fix-up' and the second bus (utilized and not) gives me the results I was originally after. Cheers Coss Link to comment Share on other sites More sharing options...
BritTim Posted October 8, 2003 Report Share Posted October 8, 2003 Thank you for taking the time to post this. It has peaked my interest and I am trying to find more information on the interactions of the different components involved, as well as the impact of the operating system. Most technical information about this available on the Internet is quite dated. The new stuff tends to be very superficial. Link to comment Share on other sites More sharing options...
adikgede Posted October 31, 2003 Report Share Posted October 31, 2003 Not directly related to your question, but related to using firewaire drives with OS X Panther. This problem seems to effect upgrades, and not new installations. http://www.wired.com/news/mac/0,2125,61031,00.html Link to comment Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.