I/O data rates using Thunderbolt 1 devices on different ports vs. being daisy chained

Discussion in 'Mac Pro' started by bxs, Feb 21, 2014.

  1. bxs macrumors 6502

    Joined:
    Oct 20, 2007
    #1
    I have three LaCie 2big Thunderbolt-1 devices that are to be connected to a new MacPro 6,1 next week. In addition I will be connecting a 5-bay Burly Port Multiplier eSATA enclosure as well.

    1) 4TB RAID-0 LaCie 2big
    2) 4TB RAID-0 LaCie 2big
    3) 6TB RAID-0 LaCie 2big
    4) 20TB Burly Port Multiplier eSATA enclosure

    The MP6,1 has 3 separate Thunderbolt-2 busses as shown below

    [​IMG]

    Ignoring the Burly device for time being I could connect the three LaCie devices to separate Thunderbolt Buses 0, 1 and 2 to configure separate data paths for each of them. If I do this I would have to connect the Burly to one of the Thunderbolt Buses being used by one of the LaCie devices.

    The Burly will be connected to the MP6,1 using a 3m Thunderbolt cable --> Sonnet Echo Express34 Adapter + eSATA Pro Expresscard 34 --> eSATA cable --> Burly. This provides full access to all 5 disks as RAID and/or JBOD, plus allows for hot-swapping.

    My question is related to daisy chaining the LaCie devices and what this may mean to expected data rates compared to them not being daisy chained.

    So let's say I use Thunderbolt Bus 1 to connect all three LaCie devices setup in a daisy chain...

    Thunderbolt Bus-1/Port-1 --> 6TB RAID-0 LaCie 2big --> 4TB RAID-0 LaCie 2big --> 4TB RAID-0 LaCie 2big

    ...and connect the Burly on another Thunderbolt Bus, such as below

    Thunderbolt Bus-2/Port-4 --> Sonnet Echo Express34 Adapter + eSATA Pro Expresscard 34 --> eSATA cable --> Burly

    This configuration leaves Thunderbolt Bus-0 completely open for Display(s)

    If i/o is being done to the LaCie devices at same time will the data rates be less than if they were each on separate Thunderbolt buses ?

    Each LaCie RAID-0 device is capable of transferring data at around 300 MBytes/sec for large file i/o. So if all three LaCie devices were moving large files at same time we have some 1 GBytes/sec to deal with. Thunderbolt-1 is rated at 10Gbps or some 1.342 GBytes/sec. From this I would assume the aggregate data rates achieved using all three LaCie devices at same time be they daisy chained or not would be the same. What are your thoughts ?

    I do not expect all three LaCie devices to be used frequently at the same time.... but want to ensure that if they are their daisy chained configuration does not impact their i/o rates in any significant way.

    Of course, I could also daisy chain the Burly off the last LaCie device as well or use the same Thunderbolt bus the LaCie daisy chain string is using and only have to use one Thunderbolt Bus for all four devices. Let's say I did this using Thunderbolt Bus-1 with the LaCie daisy chain connected to Port-1 and the Burly connected to Port-3. This would allow Ports 2, 4, 5, 6 and HDMI open for other things.

    Thanks for your time....
     
  2. JQuick macrumors member

    Joined:
    Feb 2, 2014
    #2
    Leaving all disk I/O off of the controller which serves your displays might not be necessary. This leave 20Gbit of incoming bandwidth permanently idle. If your display band width + sustained disk write bandwidth are still less than what TB2 supports, why not make use of it?

    Even if you occasionally might need to write large amounts of data over controller 0 that might be a bit slower when competing with displays, it still might make sense. Example: Archival storage or a media library are likely to be read mostly. They would be great candidates to share I/O with displays.

    I do not have a system for testing yet, so won't comment specifically on your proposed 2 controller for disk model.
     
  3. elvisizer macrumors 6502

    Joined:
    May 29, 2003
    Location:
    San Jose
    #3
    yeesh, that's a lot of lacie drives!

    I'd probably not daisy chain all the lacie's together. those are all multidrive raid devices and could in theory use a lot of bandwidth. i'd split them up as much as possible.

    your display doesn't need as much bandwidth, so it's less of a concern. probably the same with the esata chain- there's probably a controller in one of your chain of esata devices and adapters that'll throttle performance before thunderbolt will. but I'd be a little leery of using expresscard adapters in a thunderbolt chain, so I'd probably put all that on it's own bus and hope it doesn't mess with the lacie's.
    so, burly on one bus. lacie 6tb and display on the 2nd bus. and the 2 4 tb lacie's on the third bus. that's what I'd start with, but that kind of depends on how you use the storage on each of these devices, too.

    Anyway, I'd benchmark it that way for each bus, then switch up what's sharing buses, benchmark again, and continue until i'd checked all the permutations. That's a lot of devices, so testing to see what you're actually getting in the various combinations will be key!
     
  4. deconstruct60, Feb 21, 2014
    Last edited: Feb 21, 2014

    deconstruct60 macrumors 604

    Joined:
    Mar 10, 2009
    #4
    This is far more dependent upon what storage drives are inside the TB peripherals than the enclosures themselves. If the HDDs are in 180-200 MB/s range ( 1440-1600Mb/s ) which they probably are not. (that is very high for a HDD ) then 4-5 of those drives would only be driving 5-8 Gb/s of traffic. TB v1 can handle 10 Gb/s a second. Splitting less than 9-10Gb/s over multiple TB controllers isn't going to do much.

    If doing HDD accesses to files that are not monomanically focused sequential access then the effective HDD speeds are half so could have twice as many drives on a single controller.





    Doesn't matter if the disks don't deliver it. It isn't large file i/o. It is access for a single large file with nothing else going on with those disks i/o to get anywhere near the max sequential speeds that synthetic benchmarks performance put up. As soon as have to get multiple files or concurrent access will fall back closer to random access throughput speeds.


    P.S. There is no "display" Thunderbolt bandwidth consumption if just hooking up DisplayPort/DVI/HDMI devices directly to the Mac Pro. You don't have to leave just one TB controller "free' to do that. You actually do want to spread that display load around to other controllers if pushing past 2 monitors.
     

Share This Page