Need advice on RAID for editing box

[ettin]

This is going to be for a streaming camera setup (1k resolution, not webcams), and configured for post production as well.

My setup:

2 sets of 2 drives in RAID 1, or in other words, 2 seperate disks with parallel backups, 4 disks in all.

3-4 disks in raid 0 for performance in editing RAW 1k.

1 disk for the OS.



Im looking at one of these motherboards which all had some sort of built in raid controller

http://www.newegg.com/Product/Product.aspx?Item=N82E16813128362

http://www.newegg.com/Product/Product.aspx?Item=N82E16813130220

http://www.newegg.com/Product/Product.aspx?Item=N82E16813188039


Are builtin raids to be trusted in production situations? Can one raid card take care of my entire setup, or do i need a different raid card for each raid array?

[edDV]

This is going to be for a streaming camera setup (1k resolution, not webcams), and configured for post production as well.

My setup:

2 sets of 2 drives in RAID 1, or in other words, 2 seperate disks with parallel backups, 4 disks in all.

3-4 disks in raid 0 for performance in editing RAW 1k.

1 disk for the OS.


Im looking at one of these motherboards which all had some sort of built in raid controller

http://www.newegg.com/Product/Product.aspx?Item=N82E16813128362

http://www.newegg.com/Product/Product.aspx?Item=N82E16813130220

http://www.newegg.com/Product/Product.aspx?Item=N82E16813188039


Are builtin raids to be trusted in production situations? Can one raid card take care of my entire setup, or do i need a different raid card for each raid array?

Explain better your "1K" source. Is this uncompressed 1920x1080i? or what?

Explain better the type of editing you intend and software you use.

Explain the streaming requirement and codec to be used. The same machine is going to be used for HD editing and as a streaming server? Why?

Usually mother board RAID is not optimal for uncompressed HD editing.

[ettin]

its actually uncompressed 720, but ill assume i might work in 1080p for the future.
type of editing? probably the type tha involves cutting and pasting different recorded clips into a linear sequence.
I will be using premiere pro, though i might consider avid for the future.
codec will be uncompressed, as in no codec.
Same machine because it will cut costs and i will need a beefy computer to stream as well as edit, so theres no reason to build more than one.

Thanks for the mobo tip, ill will consider raid cards, but i still need more info.

[edDV]

Full-up uncompressed broadcast 10 bit 4:2:2 1080i/720p uses SMPTE-292M SDI (1.485 Gb/s) per video stream usually from and to an external RAID video server. In addition to video, SDI includes multiple 8+ uncompressed audio tracks plus metadata. You would need ~ 5 standard 7200rpm drives or ~3 fast SCSI/Raptor RAID 0 drives to support one SMPTE-292M stream.
http://en.wikipedia.org/wiki/SMPTE_292M

If you look at just a single video stream, you are looking at these kinds of data rates* (approximate)

10bit 4:2:2 1080i/29.97 ~1200 Mb/s
10bit 4:2:2 1080p/23.976 ~ 960 Mb/s
8 bit 4:2:2 1080i/29.97 ~ 960 Mb/s

720p/59.94 uses about 12% less bitrate vs 1080i/29.97 or ~ 850Mb/s for 8 bit.

850Mb/s sustained would require ~ 2-3 standard drives or 2 Raptors in RAID per stream. One should include a buffer and peak bit rate safety factor for reliability. A good hardware based RAID controller will operate independent of normal OS interrupts and be less susceptible to OS/Memory related glitches.

Most uncompressed HD cameras use SDI for interface to the computer. Aja and Black Magic make popular SDI and analog component interface cards.


* ~ 422 GB/hr for 1080i/29.97, ~ 376 GB/hr for 720p/59.94.

[ettin]

I was told this was my bandwith forumla for recording:

720x1280 (resolution) x60 (fps) x 1 (8 bit Uncompressed)*=55,296,000 or 56.2 MB/s Uncompressed RAW (i might cut it down to 30fps)

Which is ok for one caviar black per camera as it has a 56mb/s minimum write throughput

But if i stick an array in there for editing i was told that each camera stream (streamed through ethernet) becomes 3x the size when debayerd, and turns into uncompressed RGB (from its original RAW format), so i am looking at 175mb/s for uncompressed RGB 720p, which 3 of those drives in raid 0 should cover right?

[edDV]

I was told this was my bandwith forumla for recording:

720x1280 (resolution) x60 (fps) x 1 (8 bit Uncompressed)*=55,296,000 or 56.2 MB/s Uncompressed RAW

720x1280 x 60 (fps) x 1 (Byte/pixel) = 55,296,000 or 56.2 MB/s
or 56.2 MB/s * 8 bits/Byte = ~450 Mb/s
That gets you the Y luminance (monochrome) channel.

4:2:2 YCbCr has two 720x640 chroma components also one Byte per pixel.
That doubles bit rate to ~900 Mb/s. Audio, metadata plus data padding expands that out to about 960 Mb/s.

4:2:0 YCbCr has two 360x640 chroma components resulting in ~ 675 Mb/s raw data rate.

Normally effects editing is done from 4:2:2 source with internal conversion to 4:4:4 for filter calculation. 4:2:0 conversion is normally done during output encoding. If compressed 4:2:0 source is used (e.g. XDCAM/HDV), conversion to 4:4:4 is done for filtering or effects translation.

(i might cut it down to 30fps)

That might work for some applications but would appear choppy for high motion compared to 60fps.

Which is ok for one caviar black per camera as it has a 56mb/s minimum write throughput

Check this link for sustained read write specs for a Caviar Black. Single drive minimum sustained write speed is substantially less than 56MB/s. This is less of an issue for a RAID zero configuration since multiple drives run

http://regmedia.co.uk/2008/10/21/hdd_12.jpg
http://www.reghardware.co.uk/2008/10/28/review_storage_hard_drive_wd_caviar_seagate_barracude/

at average rate since drive pairs are read/written from opposite ends of the platter. So the working write rate for a pair of Caviar Blacks would be 2x 62MB/s = 124 MB/s or 992 Mb/s. There isn't much safety margin there for a 960 Mb/s stream. A third RAID drive may be necessary for non-glitch performance or for 1080i/29.97 recording.

But if i stick an array in there for editing i was told that each camera stream (streamed through Ethernet) becomes 3x the size when debayerd, and turns into uncompressed RGB (from its original RAW format), so i am looking at 175mb/s for uncompressed RGB 720p, which 3 of those drives in raid 0 should cover right?

Ethernet transfer does not require conversion from 4:2:2 YCbCr to 4:4:4 RGB but Ethernet will not support sustained 1 Gigabit/s performance even under ideal conditions. It is more like a third of that. That is why SMPTE 292M SDI coax or Fibrechannel is used for uncompressed connections of this speed. An alternative is to use lossless (e.g. Lagarith ) or low loss encoding for transfer. This adds CPU load for compression and decompression. Another option is analog component connection.
http://lags.leetcode.net/codec.html

I agree with your calculation result of three Caviar Black drives in RAID zero for 720p but not your connection method.

Questions:
What is your camera source?
What is the ouput stream destination and connection method?

[ettin]

This is my camera, the only connection it accepts is ethernet, so i assume it must work. Btw, i may have forgot to mention that this is for a sterescopic setup, so there will be two cameras working simultaneously.

http://www.sumix.com/products/cameras/smx-12a2c.html

I dont really know if the stream is 4:2:2 or 4:2:0, but i do know that the signal hasnt been debayered, resulting in 1/3 the filesize of the typical digital camera stream, where debayering happens in camera. In my instance, debayering is done in post.

Did i say caviar black? I meant caviar RE3, I got my disk bandwidth numbers from here

http://www.tomshardware.com/reviews/hdd-terabyte-1tb,2077-11.html

It says its minimum is 56MB/s, while average is around 90, but i was under the impression that the minimum is the only number that really matters. Yeah, i agree, it would be cutting it pretty close, do you think i would need an eight disk array, where each video stream would have a two disk RAID 0 mirrored by a two disk RAID 1 (is that even possible?). The thing is this streaming server will be in the same room as my production (15 foot limitation on ethernet cables for accurate sync), so i cant have 20 disks running, it would be way too loud.

Maybe i should risk stability and use velociraptors instead.

edit: forget 1080, ill never shoot in 1080, it just looks terrible anyways. I will be shooting at 720/60p and 720/30p.

[edDV]

Those SMX cameras use some sort of hardware compression to RGB for use on Gigabit Ethernet. I was talking about full uncompressed transfer and RAID storage recording as used in digital cinema and TV production. I'll read through the docs later tonight and get back to you.

[ettin]

Thanks, i appreciate the help. I have extensive experience building desktops, but this whole media server setup is getting quite complex.

I took the initiative and did some more research into RAID setups, and it occurred to me that my aforementioned idea of having a per camera setup of -> Camera -> one of two ethernet ports -> somehow telling one ethernet port to only go onto a certain raid array -> single mirrored drive, wouldnt make a whole lot of sense.

I was thinking either 0+1 or 10 would be better than running two independent 1's. The thing is ill only be using these drives for the week of production and the two weeks of post, then ill backup all files to the streaming array (the first set of 4) and replace them with new ones for the next production, so realistically my biggest issue is having a setup that will handle the bandwidth of two cameras, while preventing any catastrophic failures that would force me to call the production off.

[ettin]

Awesome, thanks for the reply!


Flake.

Goddammit, why are there so many of you.

[edDV]

You talking to me?

The SMX-12A2C is not an uncompressed camera. It uses considerable intraframe Bayer wavelet compression to get 24 bit RGB into 8bits (one byte/frame) or 40bits to 10bits. It must do more than that to get down to Ethernet rates. Decompression is CPU intensive. One source I was reading indicated minimum is a Core2 Duo plus high end display card and it wasn't clear that was real time. This tech paper questions whether near lossless can be attained real time with Bayer compression.
http://www.eurasip.org/Proceedings/Eusipco/Eusipco2006/papers/1568981744.pdf

The Sumix site has little info and none on data rates. They refer you to documentation that you need to request by phone. These guys make security/industrial cameras.

Discussion of this camera is lacking detail. One guy says Ethernet data rate is below 100 Mb/s. Another says 1080p is sent at 17Mb/s (similar rate to AVCHD H.264).
http://www.davidrubio3d.com/view_topic.php?id=462&forum_id=2&page=1
http://www.dvinfo.net/conf/alternative-imaging-methods/104870-sumix-2-3-1920x1080-cmos...tml#post997216

If you want true uncompressed 8 bit/component 1280x720p/59.94, the cheapest direct way is to get a JVC
GY-HM700U connected analog component to a Black Magic Intensity Pro feeding a 3 disk RAID zero. If you want redundant recording, you will need more drives and a higher level RAID.

I'm done. Got other things to do. Let me know what you find.

[ettin]

The SMX-12A2C is not an uncompressed camera. It uses considerable intraframe Bayer wavelet compression to get 24 bit RGB into 8bits (one byte/frame) or 40bits to 10bits.

The Sumix site has little info and none on data rates. They refer you to documentation that you need to request by phone.

Thats quite an observation for a product that has little to no information on it.

Debayering is done in post with these cameras, there is no compression or processing done to the image signal, unless your considering bayering a form of compression, but then technically there would be no such thing as a truly uncompressed camera. There is an option however to do a 2:1 signal compression, which is forced @ 1080/60 & 720/120, but optional at all other data levels. From my calculations, data rates approach 60MB/s max, and perhaps go as high as 80MB/s depending on the bitdepth, but i havent confirmed that yet.

Nm, my mistake, I shouldnt of asked you to waste your time. Thanks for the effort though.

[Cornucopia]

There's an axiom in video production: GOOD - FAST - CHEAP...pick 2!

IOW, You want FAST (aka realtime, not asynch DL), so you must choose between GOOD and CHEAP.

Those industrial cameras are good for what they're meant for: industrial/security needs. NOT cinematography.
It's not just all about the sensor, LENS/OPTICS make a huge difference here.
And with better cameras comes bigger bandwidth (it's required). Bigger BW means heavy duty RAID on the back end. And a High-performance pipeline (Dual-link HD-SDI, Fibrechannel). $$$ and more $$$. I recently finished spec'ing a stereoscopic setup for an up-and-coming graduate film school in Canada, for precisely the requirements you're talking about. And, HELL NO WAY, should you go the route your thinking.

Stereo3D specifically requires a clean signal (and as HD as you can get). With cameras that output a "compresses BW" signal, you're losing the very thing that makes stereo3d have impact--resolving power in the Z dimension (and 1080 IS better here, even if interlaced, because the Z is built by the disparity in the horizontal, hence the need for more H-rez). A little brother analogy is trying to do anaglyph stereo3d while using highly compressed 4:2:0 signal--the compromise in bandwidth and colorspace is the very thing that is destroying resolving ability of the chroma-multiplexed scheme.

I've even had a few on-line disagreements with David Rubio, specifically on his preference for these non-standard industrial cameras (especially when he doesn't want to follow DCI specs, but still expects some magical "compatibility" from these beasts).

The bottom line is: if you want to go >>>edit: Good quality<<< Stereo3D, choose:

1. SD (and here I would go with Andromeda-Modded DVX-100Bs direct to P2s or Firestores), similar to what 21st Century Video has done
or better...
2. HD (and here I would go at least HVX-200As or better pro Video/Cinema HD cameras, along with HD-SDI link to Dual-dedicated Rackmounted RAID systems)
or better...
3. Red cameras (2k, 3k, 4k, 5k) with their own proprietary RedDrives and connectors

Scott

[edDV]

The SMX-12A2C is not an uncompressed camera. It uses considerable intraframe Bayer wavelet compression to get 24 bit RGB into 8bits (one byte/frame) or 40bits to 10bits.

The Sumix site has little info and none on data rates. They refer you to documentation that you need to request by phone.

Thats quite an observation for a product that has little to no information on it.

Debayering is done in post with these cameras, there is no compression or processing done to the image signal, unless your considering bayering a form of compression, but then technically there would be no such thing as a truly uncompressed camera. There is an option however to do a 2:1 signal compression, which is forced @ 1080/60 & 720/120, but optional at all other data levels. From my calculations, data rates approach 60MB/s max, and perhaps go as high as 80MB/s depending on the bitdepth, but i havent confirmed that yet.

Nm, my mistake, I shouldnt of asked you to waste your time. Thanks for the effort though.

Maybe this exercise will help us to communicate.

Using the JVC GY-HM700U as an example of "uncompressed" 1280x720p/59.94, the camera has an optical RGB splitter and three 1280x720 CCD sensors. The direct 8bit "RAW" RGB output would be 3 x 56.2 MB/s or 168.60 MB/s (1349mb/s) or similar to SMPTE-292M.

Let's use the RED ONE and calculate back the other way. The RED ONE's REDCODE is a wide dynamic range RGB codec. In 4kx2k/24 mode it produces "RAW" 20GB/s output. Scaling that for 1920x1080p gets you ~5GB/s and for 1280x720p ~2.3GB/s. At 60fps we would have ~6GB/s (48Gb/s).

So, we have bracketed uncompressed RAW 1280x720p/60 between 1.3-48Gb/s. So if the SMX-12A2C were a true HD camera, how can it operate real time over Gigabit Ethernet? Only by applying greater than 32x compression for 720p and ~70x for 1080p.

That amount of compression is much more than HDV format although HDV is 4:2:0 not 4:4:4.

The way "RAW" is used for these cameras means uncorrected, not uncompressed.

Bayering is a form of 2:1:1 data compression since green is sampled twice the rate of blue or red.

Also, debayering in post means that for real time playback or >1x timeline scrubbing, the RAID needs to be sized for full uncompressed 4:2:2 or 4:4:4 720p.That works out to 3 to 5 drives. If you transcode to Cineform, you should be able to edit from a single drive.

[MattMiller]

The SMX 12A2C is not sending raw 4:4:4 data through the gigabit connection. Bayer reconstruction occurs on the PC side and hence the only data that is being transferred is 12-bit monochromatic or a brightness level of 0-4095 per pixel. Your math assumes that color information is being transferred when it is not. Along with data compression, this is how all of that information fits into a 1Gb/s stream.

178MB/s at 30P if you're referring to the uncompressed AVI that the software generates. A 4-way RAID 5 (hardware) would suffice for editing. You could also transcode everything into cineform.

Regarding the sensor and optics, AlatSens makes some of the world's best sensors. On the contrary, they are excellent for cinema use. There are various adapters that allow the use of professional PL lenses and hence the best optics can be used on this camera head. If you know what you're doing with the color processing, you'll get a more film-like image out of this camera than you'll get out of a RED.

[Cornucopia]

According to sumix:

• @1080, only does 30fps (not 60!) and only at 8bit
• comes with camera controller/capture app that is meant for industrial apps - you want cinematography, you got to create your own app
• doesn't sync externally, so can't do 23.976
• no audio at all
• only 2/3" chip, so certain low-light-summed captures (adjustable iso) and short DOF options not easy
• need pc attached for it to work

And if you google "digital cinematography smx 12a2c" (and filter out this thread and sumix's own material), you get about 2 threads, one in DVXUser where they come to the conclusion that's it's "not quite ready for prime time" and users have moved on to a Red or SI of some sort. Not my idea of a cinematography camera. Yet major motion pictures are already being show using the Red and SI cameras. Like I originally said...

Scott

[MattMiller]

Which is precisely what we've been doing for 10 months. The camera does 60p at 1080 once you get all the bugs worked out and only on a particular chipset.

23.976 can be achieved through our interface.

Audio can be acquired through the system's sound mapper. We're using 24-bit, 48KHz w/ 102dB SNR

2/3" DOF at F2.0 is identical to 35mm DOF at F5.6. Low-light isn't as great as a 35mm sensor but remember that the RED can barely call itself 320asa.

Fixed pattern noise is still an issue but there are ways around that.

Our entire rig fits on your shoulder - PC included.