Not long ago, Macworld put the new Mac Pro to the test on the front of connectivity and external drive performance. At one point the test involved 36 devices connected via Thunderbolt, USB 3, and even FireWire 800 (via Thunderbolt docks), including an extensive list of external hard drives plus an Apple Thunderbolt display, two Apple Cinema Displays, and a 27-inch HP LED display. The results shed very interesting light on some of the capabilities -- and possible limitations -- of Thunderbolt daisy chains.
Macworld gathered a veritable cornucopia of devices to hook into the little trashcan-looking workhorse, including external drive units from Buffalo, G-Tech, LaCie, OWC, Promise, Seagate, Toshiba, and Western Digital. At the point where 36 devices were chained to the Mac Pro, external storage totaled over 100TB. One of the most interesting findings was that, apparently, an external drive array's placement in a Thunderbolt daisy chain can degrade its throughput, even if devices in between aren't doing anything. Check out Macworld's great write-up for the complete list of attached devices, and the full test results. Also see below (click for larger version):
Here's Macworld on some of its methodology and details (my emphasis):
When running a script that copied data from the Mac Pro’s internal PCIe connected flash storage to each of the drives, the combine power draw peaked at a whopping 865 Watts. The script also tracked the amount of time it took each drive to write 6GB of data. The fastest was OWC’s Mercury Helios, which was able to write at an average of 271 MBps with all other drives running. In second place was the Promise Pegasus2 R6, which wrote the data at an average of 199 MBps. Bringing up the rear were an unlucky few drives connected to the USB 2.0 ports of an Apple Cinema Display. They only managed to write at 3 MBps.
According to the OS X activity monitor, Macworld saw "overall throughput of around 3 GBps [24Gbps] with all the drives running." One can make a few reasonable assumptions about daisy-chaining devices with Thunderbolt 2: it's probably best to place Thunderbolt 1 devices at the end of a chain, for instance, because they in themselves and in 'loop-through' are likely limited by their older controllers; external drives will be limited by whatever bottlenecks they're inherently subject to; and, the performance of a whole chain may vary depending on the at-the-moment activity of each the devices or drives in the chain and/or other chains, exemplified above with the Pegasus' performance as compared to the figures to follow.
To me, this was the real kicker -- the performance of a device may be affected by where it it's physically placed in the daisy chain, regardless of the activity (or lack thereof) of devices in-between. Case in point, the Pegasus R6 RAID array Macworld played in the tests apparently averaged 709.8 MB/s write and 584.7 MB/s read both at the front of a 6-device chain and as the only external volume mounted -- but, when moved down the line to the end of its chain, the Pegasus dropped to 556.7 MB/s write and 591.5 MB/s read (the latter of which is comparable to the earlier read-speed figure).
As Macworld makes no indication to the contrary, I'm assuming the Pegasus was the only device doing heavy lifting in the chain-placement round of tests. This would seem to indicate that some degradation of device performance can be associated with placement 'at the back of the line.' Though it probably won't constitute a deal-breaker for many users, it's certainly good to know.
Be sure to check out Macworld's full article on the subject -- and if you're interested in the matter further, you might enjoy Tom's Hardware's "How Fast Is Thunderbolt, Really? Tom's Tests With 240 TB" and Macworld's earlier "Thunderbolt: How devices affect each other on a daisy chain." These were written before the advent of Thunderbolt 2, but are still good reads.