[Michael Tiemann]

One small note: the QNAP and the OS absorbs quite a bit of storage from the raw disks to make its RAID volumes. With 8x 12TB disks you’d think you get a 72TB RAID6, but after formatting and allocation, it’s just shy of 65TB. I all it 64TB because I like round numbers. That doesn’t change anything Bob says about the HW, but it might change your $/TB equations as you look at piles of unmanaged drives vs. A tidy NAS RAID.

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

RAID is not backup. It is so not backup that there’s a famous thread you can read about it:

https://serverfault.com/questions/2888/why-is-raid-not-a-backup

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

Sure! Sent you a reply to your email.

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

Well, it looks like I get to follow up on my own question(s).

TL;DR: the 872XT is quite fast when using the thunderbolt bridge to connect to my late 2017 iMac Pro. The Blackmagicdesign Disk Speed Test reports over 1000MB/sec write speed and over 1400MB/sec read speed, the latter being fast enough to play 2160p60 10 Bit YUV 4:2:2 faster than real-time (and write it at about 44 fps). The XT is what makes the 872XT worth its premium.

The 10Gbe network speeds are fast, but not blazingly fast. I fault Mojave, mostly, for that.

Here’s a look under the hood at some raw numbers, first from the storage side:

[~] # qcli_storage -t
fio test command for LV layer: /sbin/fio --filename=test_device --direct=0 --rw=read --bs=1M --runtime=15 --name=test-read --ioengine=libaio --iodepth=32 &>/tmp/qcli_storage.log
fio test command for File system: /sbin/fio --filename=test_device/qcli_storage --direct=0 --rw=read --bs=1M --runtime=15 --name=test-read --ioengine=libaio --iodepth=32 --size=128m &>/tmp/qcli_storage.log
Start testing!
Performance test is finished 100.000%...
VolID VolName Pool Mapping_Name Throughput Mount_Path FS_Throughput
1 DataVol1 288 /dev/mapper/cachedev1 1.10 GB/s /share/CACHEDEV1_DATA 723.16 MB/s


[~] # qcli_storage -T
'fio test command for physical disk: /sbin/fio --filename=test_device --direct=1 --rw=read --bs=1M --runtime=15 --name=test-read --ioengine=libaio --iodepth=32 &>/tmp/qcli_storage.log
fio test command for RAID: /sbin/fio --filename=test_device --direct=0 --rw=read --bs=1M --runtime=15 --name=test-read --ioengine=libaio --iodepth=32 &>/tmp/qcli_storage.log
Start testing!
Performance test is finished 100.000%...
Enclosure Port Sys_Name Throughput RAID RAID_Type RAID_Throughput Pool
NAS_HOST 1 /dev/sdg 235.72 MB/s /dev/md1 RAID 6 1.21 GB/s 288
NAS_HOST 2 /dev/sdh 240.79 MB/s /dev/md1 RAID 6 1.21 GB/s 288
NAS_HOST 3 /dev/sdf 239.00 MB/s /dev/md1 RAID 6 1.21 GB/s 288
NAS_HOST 4 /dev/sde 252.49 MB/s /dev/md1 RAID 6 1.21 GB/s 288
NAS_HOST 5 /dev/sdd 236.57 MB/s /dev/md1 RAID 6 1.21 GB/s 288
NAS_HOST 6 /dev/sdc 253.76 MB/s /dev/md1 RAID 6 1.21 GB/s 288
NAS_HOST 7 /dev/sdb 240.28 MB/s /dev/md1 RAID 6 1.21 GB/s 288
NAS_HOST 8 /dev/sda 241.76 MB/s /dev/md1 RAID 6 1.21 GB/s 288


Those numbers give us the “guaranteed not to exceed” capabilities of the disk array.

Now, looking at the network fabric…the Thunderbolt bridge:

michael$ iperf3 -c 169.254.9.151
Connecting to host 169.254.9.151, port 5201
[ 5] local 169.254.91.32 port 50480 connected to 169.254.9.151 port 5201
[ ID] Interval Transfer Bitrate
[ 5] 0.00-1.00 sec 1.96 GBytes 16.9 Gbits/sec
[ 5] 1.00-2.00 sec 1.88 GBytes 16.1 Gbits/sec
[ 5] 2.00-3.00 sec 2.05 GBytes 17.6 Gbits/sec
[ 5] 3.00-4.00 sec 2.00 GBytes 17.2 Gbits/sec
[ 5] 4.00-5.00 sec 2.00 GBytes 17.1 Gbits/sec
[ 5] 5.00-6.00 sec 2.04 GBytes 17.6 Gbits/sec
[ 5] 6.00-7.00 sec 1.98 GBytes 17.0 Gbits/sec
[ 5] 7.00-8.00 sec 2.10 GBytes 18.1 Gbits/sec
[ 5] 8.00-9.00 sec 2.14 GBytes 18.4 Gbits/sec
[ 5] 9.00-10.00 sec 2.05 GBytes 17.6 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate
[ 5] 0.00-10.00 sec 20.2 GBytes 17.4 Gbits/sec sender
[ 5] 0.00-10.00 sec 20.2 GBytes 17.4 Gbits/sec receiver


And the 10G ethernet connection:

michael$ iperf3 -c 192.168.1.19
Connecting to host 192.168.1.19, port 5201
[ 5] local 192.168.1.41 port 50960 connected to 192.168.1.19 port 5201
[ ID] Interval Transfer Bitrate
[ 5] 0.00-1.00 sec 964 MBytes 8.08 Gbits/sec
[ 5] 1.00-2.00 sec 904 MBytes 7.59 Gbits/sec
[ 5] 2.00-3.00 sec 979 MBytes 8.21 Gbits/sec
[ 5] 3.00-4.00 sec 1000 MBytes 8.39 Gbits/sec
[ 5] 4.00-5.00 sec 1.07 GBytes 9.13 Gbits/sec
[ 5] 5.00-6.00 sec 388 MBytes 3.27 Gbits/sec
[ 5] 6.00-7.00 sec 933 MBytes 7.82 Gbits/sec
[ 5] 7.00-8.00 sec 1.06 GBytes 9.10 Gbits/sec
[ 5] 8.00-9.00 sec 1.06 GBytes 9.15 Gbits/sec
[ 5] 9.00-10.01 sec 931 MBytes 7.77 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate
[ 5] 0.00-10.01 sec 9.15 GBytes 7.85 Gbits/sec sender
[ 5] 0.00-10.01 sec 9.15 GBytes 7.85 Gbits/sec receiver


I’ll point out a few things in the above: the first is that the results vary in just that run from over 9Gbps to barely 3Gbps. I’ve occasionally see less than 1Gbps show up, and sometimes less than 1Mbps! But generally it tends to be mid-7 to mid-8 going from the iMac Pro to the QNAP. Reversing the flow (with -R option), the numbers drop about 1.5-2Gbps. That’s right: the Mac is faster when it’s writing than when its reading over its 10Gbe port.

Not so a MacBook Pro using an OWC TB3->10Gbe adapter. That unit (which looks like a re-branded Sonnet Solo) consistently maintains over 9Gbps each way (when running Mac-to-Mac over a QNAP 10Gbe switch). Since both the iMac Pro and the MacBook Pro are running Mojave 10.14.2, I would say that Apple has done something wrong that’s hurting their 10Gbe port speeds.

Running the disk speed test, write speeds from the iMacPro tend to be well over 700MB/sec (enough to write 2160p30 10bit YUV 4:2:2 in real-time) and just under 600 MB/sec (enough to read 2160p24 10bit YUV 4:2:2 but often not quite enough to do 2160p30). And this asymmetry is no surprise because we know that our network speeds are similarly asymmetric.

I will also say that when running iperf3 with the -P option, I can typically fully saturate the 10Gbe connection with -P4 (four parallel readers or four parallel writers), which means that it could theoretically serve about 8 layers of 8K R3D media at REDCODE 8:1 at 24p.

But the thunderbolt interface is faster, more symmetric, and more consistent in its delivery of bandwidth.

P.S. The 12TB drives format down to 10.91TB after QNAP takes their chunk, and they format to a RAID6 with 64.8TB worth of capacity. Formatted capacity = 90% of unformatted capacity.

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

I’m afraid the forum software messed up the link you posted.

BTW, I have an 872xt arriving shortly. I will be loading it with 8x WGST He12 drives and connecting it to a QNAP QSW-1208-8C-US 10GbE switch. I wonder how many days it will take to build the RAID.

The first project I’m loading up has will have about 25TB of R3D media and close to 30TB of media overall. I already have 14TB of R3D media and over 17TB overall. But I still have more to shoot in early January. I then have two other projects that will shoot about 10TB each before the end of January.

The QNAP will be working hard right off the bat!

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

I suck. I mis-remembered the product number, which is actually the 872-XT: https://qnapdirect.com/collections/bays/products/qnap-tvs-872xt-i5-16g-8-bay-diskless-nas?variant=14679133454403

I tried to edit the post to correct the product number, but I don’t have sufficient privilege to do so. Sorry to sow confusion, and I take your point about needing to get hands-on experience before saying anything constructive.

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

I would like to know this, too! Could it be that the answer is “Use the $30,000 Panel” because that panel provides a level of picking precision that’s impossible to achieve using the mouse or limited numerical input precision?

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

Ah! Now I see!! That’s the trick I needed.

What messed me up was that when I set the Path Mask, all the handles of the mask disappeared. Which confused me, because most times you set a keyframe, all the handles, objects, etc., remain visible. It is true that the effect of the mask does remain when setting a Mask Path. But again, I was fooled by the fact that the mask outline and handles all disappeared. I did find that when I moved the transport and re-selected the overall mask, the handles reappeared, allowing me to move the mask, and set a new keyframe, whereupon all the handles and the mask outline disappeared. But the effect of the mask, now moved, gave me the effect I wanted.

THANK YOU!

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

Thanks for the response, but your answer makes no sense to me. There’s nothing moving in the image for Premiere Pro to track. I want to be the one to say “first, go here, then go there”. If I can create some kind of default track and then force it to track from one location to another, I’m happy. But if masks cannot move but for the magic of the tracker finding something to track, then that’s not a solution for me.

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/

[Michael Tiemann]

I take it back…mask expansion is a great answer, but only in the case that the mask is relatively near the center of the frame. With a mask expansion of -220, I can shrink the mask down to the size I want, and with an expansion of 900 I can get it to expand past the edges of the frame. But if the mask is near the edge, not near the center, the 1000 max value of mask expansion is not enough to expand past the far edge of the frame. (It barely expands past the center, to be honest.)

If I could animate the center point of the mask, that would solve my problem. But I don’t see any way to do that.

Manifold Recording
Pittsboro, NC
https://manifoldrecording.com/