Crazy speeds with three TB2 externals and a 2013 Mac Pro

Discussion in 'Mac Pro' started by macguy18, Jul 30, 2014.

  1. Pressure macrumors 68040

    Pressure

    Joined:
    May 30, 2006
    Location:
    Denmark
  2. AidenShaw, Jul 30, 2014
    Last edited: Jul 30, 2014

    AidenShaw macrumors P6

    AidenShaw

    Joined:
    Feb 8, 2003
    Location:
    The Peninsula
    #3
    Recipe for disaster

    Running RAID-0 with multiple enclosures, multiple power supplies and multiple non-latching TB-2 cables is very risky.

    You'd probably get a lot of practice at restoring everything from backups....
    ______

    On the other hand, if you need a fast input disk, a fast output disk, and a fast scratch disk and your final result will be moved to a safe array - then a RAID-0 array per enclosure would be sweet.
     
  3. iBug2 macrumors 68040

    Joined:
    Jun 12, 2005
    #4
    It seems just like yesterday when some people were arguing about whether or not this would be possible with the new Mac Pro. :)
     
  4. bxs macrumors 6502

    Joined:
    Oct 20, 2007
    #5
    I have 3x LaCie 2big TB1 devices. Each one is 6 TB and each one is RAID-0. Each one was bought off eBay at around $300 ea. They all have around 1 to 2 years left on their original LaCie 3 yr warranty.

    Using Disk Utility I RAID-0 across these three LaCie units and observed around 1200 MB/sec on my 2013 MP6,1 using three separate TB2 bus ports.

    Works like a charm.

    Once setup cable latching is of no concern... and yes there are several points for failure, but so far no problems... fingers crossed.

    As with anything these days, backups must be done as you just never know...
     
  5. VirtualRain, Jul 31, 2014
    Last edited: Jul 31, 2014

    VirtualRain macrumors 603

    VirtualRain

    Joined:
    Aug 1, 2008
    Location:
    Vancouver, BC
    #6
    Does anyone understand how the HD benchmark is possible? What am I missing here? From what I can gather, they are using a RAID0 array of 12 drives (3TB each). If each drive has an STR of 200MB/s, the combination would only yield 2400MB/s. Yet they are achieving 4000MB/s, which would imply that each 3TB drive has an STR of 333MB/s. WTF drives are these? I want some! :)

    The SSD benchmark is actually disappointing. If each chassis has 4 SSDs that should provide about 2000MB/s per cabinet... or 6000MB/s combined. So either the cabinet is not able to fully saturate the TB2 bus, or the Mac Pro is unable to handle 3 fully saturated TB2 buses. I'm guessing the former.
     
  6. paulrbeers, Jul 31, 2014
    Last edited: Jul 31, 2014

    paulrbeers macrumors 68040

    Joined:
    Dec 17, 2009
    #7
    You make a very good point about the Mechanical drives.... The drives used (per OWC) in their Thunderbay 4 are Toshiba DT01ACA. Searching for benchmarks, the BEST results I could find were for their 1TB drives (the benchmarks for 2 and 3 were worse than these results) resulted in a Sequential Read of "only" 180GB/s (approximately). 12 x 180 = 2160MB/s if it was perfectly balanced across all 12 drives. Something is definitely wrong here.....

    Now for the SSD's, I am not surprised. 1400MB/s seems to be about the right bandwidth for 4 SSD's in a RAID box (here's a review for the Pegasus 2 using 1TB EVO's http://www.tekrevue.com/thunderbolt-2-benchmarks-promise-pegasus2-r4-four-samsung-evo-ssds/). So figuring 1400 x 3 = 4200 and you probably lose a little off the top due to RAIDing 3 external boxes, I'd say 3900 makes perfect sense.

    Edit: They have the "allow cache effects" turned on. I'm going to guess that the speed is coming from the Cache only of the drives which would explain how it blew past the sustained drive speeds....
     
  7. VirtualRain macrumors 603

    VirtualRain

    Joined:
    Aug 1, 2008
    Location:
    Vancouver, BC
    #8
    Yeah... since I posted this, I noticed they are allowing cache effects (see your "Edit" now too). Certainly not a best practise for benchmarking storage, but that would explain the inflated results for the HDs. It pretty much makes the article useless IMHO.

    As for the SSDs... I see what you're saying, but unfortunately, we don't know if these boxes are performing to the same level as the Pegasus since they've also enabled cache effects on that test as well.
     
  8. wonderspark macrumors 68030

    wonderspark

    Joined:
    Feb 4, 2010
    Location:
    Oregon
    #9
    Haha, sure, ok!
    With my cache enabled, I get 3633MB/sec read speeds, too!
     

    Attached Files:

  9. paulrbeers macrumors 68040

    Joined:
    Dec 17, 2009
    #10

    Certainly not a best practice at all. Great for marketing! ;)

    ----------

    Just proves how much the Cache can have an affect on the benchmarks eh?
     
  10. bxs, Jul 31, 2014
    Last edited: Jul 31, 2014

    bxs macrumors 6502

    Joined:
    Oct 20, 2007
    #11
    I agree that OWC should have highlighted the fact they were allowing 'cache affects'. They did at least show that the "Allow Cache" check box was set in their report... so they really were not hiding this fact.

    In reality, the the cache is employed for applications in most cases unless direct i/o is specifically being performed. So, I see the results as being useful.

    Just for the record, I ran AJA on my MBP8,3 that has an empty TB1 LaCie 6TB RAID-0 enclosure connected. I unchecked "Disable file system cache" and ran a quick test... and it displayed Write at 275 MB/s and Read at 4,303 MB/s. :eek:
     

    Attached Files:

  11. iBug2 macrumors 68040

    Joined:
    Jun 12, 2005
    #12
  12. Pressure macrumors 68040

    Pressure

    Joined:
    May 30, 2006
    Location:
    Denmark
    #13
    Sure but having 36TB compared to 3TB of space for the same price might make a difference for some.
     
  13. iBug2 macrumors 68040

    Joined:
    Jun 12, 2005
    #14
    Well, size vs stability. It's much safer to do this with only 3 drives.
     
  14. AidenShaw macrumors P6

    AidenShaw

    Joined:
    Feb 8, 2003
    Location:
    The Peninsula
    #16
    ...and maybe someday they'll post real-world benchmarks showing what applications actually achieve.

    I'm so tired of that "Black Magic" 'binocular' graphic showing large I/O sequential numbers - which misses completely the real world apps that have a mix of read and write and a mix of small transfers and large transfers.

    But, the Black Magic test has a nice shiny graphic, and we know how important "shiny" is.

    OWC is pretty shady here - based on this thread alone I would never buy anything from them. Not clearly explaining that their numbers involve host caches (not completely hiding that fact - but not highlighting it) is dishonest and misleading.
     

Share This Page