The Pegasus2 M4: Performance

Let's start with single-drive performance to get the baseline level of performance so we have an idea of what to expect from a 4-drive array. We ran HD Tach on one of the 1TB 5,400rpm Toshiba drives as HD Tach produces a nice graph of performance across all LBAs. 

Peak performance is 120MB/s for one drive, so it is reasonable to expect a maximum performance of ~480MB/s from a RAID 0 array. However, by the last LBAs the throughput drops to half that amount.

For array performance, we use Iometer due to its flexibility. Sequential performance is tested with a transfer size of 2MB at queue depth of one and the test runs for one minute, while 4KB random performance is tested at queue depth of three for three minutes. All tests are run with an 8GB LBA space, so we are looking at the best case performance here – the HD Tach graph above gives you an idea of how the performance degrades as the array is filled.

Promise Pegasus2 M4 4TB (4x1TB) Performance
  2MB Sequential Read (QD1) 2MB Sequential Write (QD1) 4KB Random Read (QD3) 4KB Random Write (QD3)
Pegasus2 M4 (RAID 0) 480.6MB/s 466.81MB/s 0.95MB/s 4.42MB/s
Pegasus2 M4 (RAID 5) 355.5MB/s 356.0MB/s 0.94MB/s 0.67MB/s
Pegasus2 M4 (RAID 10) 408.4MB/s 240.5MB/s 1.11MB/s 2.15MB/s

In a RAID 0 configuration, the M4 manages up to 480MB/s, which is where the hard drive becomes the bottleneck. That is far from the maximum performance of Thunderbolt 2 (20Gbps or 2.5GB/s) but over 100MB/s faster than what USB 3.0 is typically capable of providing.

With RAID 5 the write performance takes a hit because of the parity writes (you can only write data to three disks simultaneously as the fourth one will be writing parity data), but read speed is also affected. It could be a limitation of the RAID controller itself because some RAID controllers do not tend to work as well with RAID 5 because of the processing power required for parity calculation. RAID 10 also experiences similar performance loss as in theory RAID 10 should provide the same read performance as RAID 0, but that is not the case with the M4 and its RAID implementation. 

I wanted to see what the M4 is capable of when fitted with faster hardware, so I took four 256GB SSDs and configured them in a RAID 0 array. I ran the same tests as above but added a test with queue depth of 16 to see the maximum throughput of the M4.

Promise Pegasus2 M4 1TB (4x256GB SSD) Sequential Performance
  2MB Sequential Read (QD1) 2MB Sequential Read (QD16) 2MB Sequential Write (QD1) 2MB Sequential Write (QD16)
Pegasus2 M4 (RAID 0 - SSD) 579.1MB/s 719.0MB/s 532.6MB/s 617.0MB/s
Pegasus2 M4 (RAID 0 - HDD) 480.6MB/s 478.4MB/s 466.81MB/s 456.4MB/s

With SSDs the M4 takes more advantage of Thunderbolt, although performance is still not that great. Four SATA 6Gbps SSDs should be capable of ~2GB/s in sequential read, so it is clear that the SATA/RAID controller (or the RAID controller) is limiting performance. For mechanical storage that is not an issue but we can see why Promise is not offering an SSD configuration – it simply would not be worth the extra cost since the performance upgrade is rather insignificant, at least for video work where random accesses are not critical.

Promise Pegasus2 M4 1TB (4x256GB SSD) Random Performance
  4KB Random Read (QD3) 4KB Random Write (QD3)
Pegasus2 M4 (RAID 0 - SSD) 16.75MB/s 30.1MB/s
Pegasus2 M4 (RAID 0 - HDD) 0.95MB/s 4.42MB/s

Random performance is obviously multiple times better with SSDs but nowhere near what it should be with four SSDs in RAID 0. Even a single SATA 6Gbps SSDs scores around 90MB/s in random read and easily over 200MB/s in random write, so the SATA/RAID controller appears to be the bottleneck once again.

The Pegasus2 M4: Software Final Words
POST A COMMENT

39 Comments

View All Comments

  • Zak - Friday, September 12, 2014 - link

    The 4xSSD performance is a little disappointing indeed. That would be #1 reason to get this enclosure. I get faster speeds, over 1GB/s reads, out of two SSDs on the onboard Intel RAID controller. Reply
  • simonrichter - Friday, October 3, 2014 - link

    I agree, it is rather disappointing and it makes it an average storage devices that cannot match up to the top ones on the market (for example http://www.consumertop.com/best-computer-storage-g... ). But it should be interesting to see if they release an updated version of it. Reply
  • jonb8305 - Wednesday, October 15, 2014 - link

    Promise has an SSD version of the m4 with way better performance than what was stated here. Reply
  • bill.rookard - Friday, September 12, 2014 - link

    While I do somewhat see the use of a device like this, I'm not sure I see it really serving any real niche effectively.

    It's made to use small, portable drives, but it's not portable as it requires external power.
    It should be quick, but it's limited by the internals to about 1/4 of it's theoretical top speed.
    It uses the expensive Thunderbolt interface to be fast, but again, it's limited internally.
    It offers four drives, but keeps them to 2.5" drives without making the unit truly portable.
    It offers four drives for capacity, but then only offers 1TB drives.

    This device just seems like a whole series of compromises without really SERVING a niche effectively.
    Reply
  • JohnMD1022 - Tuesday, September 16, 2014 - link

    Why not offer it as a bare box? Reply
  • Drizzt321 - Friday, September 12, 2014 - link

    How about running the SSDs as single disks and using Windows RAID to check performance. Cut out the RAID controller, which them will leave us the SATA controller to test that to see if it's the RAID controller, or the SATA controller.

    And I agree, it's too bad it can't be bus-powered. Maybe when USB3.1 with Type-C connectors comes along it'd be able to power something like this. 100W is quite a bit of power, especially with 2.5" drives!
    Reply
  • repoman27 - Sunday, September 14, 2014 - link

    If Kristian was correct in his guess that there's a PMC-Sierra PM8011 lurking under that heat sink, which is quite likely seeing as Promise uses that chip in several other products already, it's an RoC (RAID-on-Chip). So the SATA (actually SAS in this case) controller and the RAID controller aren't terribly separable.

    The performance scaling actually looks damn near perfect with the HDDs, and indicates that the RoC is actually a beast for the intended workload. I'm not sure why Kristian thought RAID 5 read performance would be higher. If you only stripe across three drives and write parity data to the 4th, it would be pretty challenging to read back faster than 3x the maximum a single drive can muster. In this case, the Pegasus2 M4 hit 355.5 MB/s vs. 120 MB/s for a single drive, or near as makes no difference 3x. And the 15% performance hit for sequential reads in RAID 10 doesn't seem too egregious, especially seeing as random reads went up by almost 17%.

    Kristian never mentioned what he was using for SSDs or if they were all identical. I'm guessing whatever he used, the RoC simply wasn't tuned for it. Although who knows, maybe the same test on a Mac would have yielded radically different results.

    100 W may be a lot for 4x 2.5-inch HDDs, but the Pegasus2 M4 appears to be packing a compact internal 110 W PSU from FSP. That's nuts!
    Reply
  • HigherState - Friday, September 12, 2014 - link

    I know they say performance on win vs mac should be close, however those numbers are so dissapointing that its possible that its os driver related as well. Someones bound to have an old MBP to lend you for the test Reply
  • repoman27 - Sunday, September 14, 2014 - link

    Well, the performance of the unit as it shipped isn't really disappointing at all. I mean, aside from not using something slightly peppier than the Toshiba drives, like maybe HGST Travelstar 7K1000's, what was Promise supposed to do?

    Clearly the SSD experiment was performed with a set of drives that had in no way been validated against Promise's firmware.
    Reply
  • The_Assimilator - Friday, September 12, 2014 - link

    Or you could spend $700 on four 7200rpm 4TB 3.5" drives and a cheap RAID controller card, and build a RAID-10 setup that also wouldn't be portable, but would be fast and have 8 times the capacity of this POS. Reply

Log in

Don't have an account? Sign up now