This website is released under
On the Bright Side of Data Migrations
Let’s be very clear from the very beginning: I do not consider data migration a good thing at all for the archive community. On the contrary, it costs a lot of time, money and effort to be achieved accurately. But it cannot be avoided. I will discuss here how data migrations can be used efficiently for modifying, where necessary, the archive’s containers, codecs, data and metadata. During the two dozen of data migrations we have carried out for ourselves and our clients, we could actually fix errors in the structure and metadata of the archive, and also we could replace obsolete or endangered formats with current ones. This allows us to change or adjust the strategy when needed. We could update the data and therefore realise maintenance of the digital archive.
In the Jungle of File Formats
When I announced that I was going to present on the bright sides of data migration,1 my colleagues replied to me that it will be a very short speech… five or ten seconds at best! And, indeed, data migration is mostly a very bad thing, also because it’s imposed by the vendors on the archives and it costs a lot of time and money that archives are usually lacking of anyway. There is only a narrow time frame in which the archive has to act, in order to take benefit of affordable costs. Inaction is almost never a good choice in the digital domain. Often it’s better to choose an intermediate step, one that can be improved or modified later. Nevertheless, my presentations – and consequently this article – do focus on how a data migration can be used in a positive way, for example to modify where needed the archive’s containers and codecs. And I do apologise that my text has a strong personal tone.
In 2002 I began teaching the conservation and restoration of moving images at the Bern University of Applied Sciences. Since that first lecture, every single year I am explaining to the students that in the digital domain it’s really essential to preserve not only the files, but also at least the source code of the codecs used to create those files. This remains good practice 16 years later, in my opinion. Of course, it’s easier when that code is released as open source. Otherwise it may be much more difficult or even impossible, because one would need to steal intellectual propriety and/or to reverse engineer to find the applied algorithm. It’s completely feasible to crack the code, as it has been done for example for all the ProRes 422 and 4444 variants, but again it costs time and money to the archive. In my opinion, this is indeed a very important point: we as an archival community do need to know in every single detail the file formats we are supposed to preserve. Not each digital archivist needs to be a technical nerd, but at least one person in each team should be highly skilled.
HuffYUV and FFV1
When I left my position as head of preservation at the Cinémathèque suisse in 2003, I established my own conservation and restoration company, AV Preservation by reto.ch, opening both a full photochemical lab and a full digital lab. As at that time no scanner was available so as to be able to gently handle fragile archival material, my team and I often had to build our own equipment, for example by modifying a Truca-like film printer, and sometimes by manufacturing from scratch the machines we needed. Back in 2003 the only possibility I had to encode, at a reasonable speed, lossless video during the digitising process was by using the HuffYUV video codec.2
Some months later I also started testing FFV1, but I have to admit, that at that time I used more the HuffYUV encoding rather than the FFV1 encoding, because it was a more mature video codec and still today it’s the faster one of the two.3
We did use the AVI container, because at that time it was the easiest for me to implement in our workflow. And I have to admit that at that time I did not know of the existence of Matroska yet, which has been born one year earlier, in 2002.
TIFF, DPX and OpenEXR
For the single-image based film content we mainly used, since 2005, TIFF in folders and DPX in XML containers.
In 2013 we started testing OpenEXR and the year after we presented for the first time how the Academy’s ACES and OpenEXR can be used in an audio-visual conservation and restoration context. Since spring 2015 our regular digital restoration workflow, which applies to two thirds of our work, is based on these technologies. We do actually use MXF, and today we do it in conformity with AS-07. We have presented our experimentations during a Late Summer School, in which we mixed together both high-end photochemical experimentations and high-end digital experimentations.
Fifteen years later, in 2018, we are still carrying out in-depth experimentation and our test implementations on both hardware and software are going in various directions.
The majority of the current mid-range and low-range film scanners use a Bayer pattern sensor. This is not a crime at all. On the contrary, it’s often the only possibility that archives can afford. The open source codec CineForm RAW or the new proprietary codec ProRes RAW allow users to store these data forms natively. Today FFV1 cannot store raw Bayer-encoded video and can only store the image content after the additional step of de-Bayering, also called de-mosaicing. FFV1 should implement in version 4 the possibility to handle natively Bayer-based content as well.
Currently FFV1 is limited to the Y′CBCR and RGB pixel-format families. There is an increasing need to store multi-spectral content from moving image as well, as done for many years in other fields of conservation and restoration.4 In one experimental implementation, for example, we apply a quick-and-dirty hack by storing 15 spectral scans as 5 «RGB» images. This has many similarities to the Filmic project by Jim Lindner, and some similarities as well to high-dynamic range (HDR). Our approach allows us to store in a transparent way this data today. I am confident that once an open-source file format for multi-spectral moving images is established, we will be able to trans-multiplex and/or transcode our data in that format.
The research on 24 bit per colour channel is going on both the film industry in Hollywood and the IT and game’s industry in Silicon Valley. Of course, at some point such files will also arrive in the archives and therefore we are experimenting on those formats as well and are testing their appropriate handling.
In the internal context of my company, the adoption of a format in the outside world is not an argument and internally we do use many experimental formats on a daily basis. We simply apply at the end a script or program to generate standard files for the delivery to the clients.
In 2014 we migrated our internal archive from LTO-4 cartridges, recorded in proprietary formats and as old-styled TAR archives, to LTO-6 with the more modern LTFS formatting. That was 5.7 PB of data on almost 7500 LTO-4 cartridges. En passant, we also transcoded all the uncompressed Y′CBCR 4:2:2 video files into lossless compressed FFV1 and wrapped it into a Matroska container. That has saved between one and two thirds of the required storage, depending on the image content. The whole process resulted with an archive of less than 2000 LTO-6 cartridges.
Then we completed two dozen data migrations for different clients’ archives. During almost all those data migrations we actually have fixed errors in the structure of the archive, or in the naming and the metadata of the files, and sometimes we have transcoded obsolete formats into current ones as well. Therefore we actually have maintained and even updated the archives. For example, in one case, the client wished us to replace the existing MD5 checksums with newer SHA-1 checksums.
Let’s take a look at the workflow. The files are read from the source cartridge. The content is sent to a script or a computer program, which can modify what needs to be modified.5 Then the writing procedure writes the files onto the destination cartridge. The external script or computer program can do a variety of actions on the data, including modifying the container and/or the codec. It might be the audio codec and/or the video codec (and/or other codecs as well). This feature opens a lot of possibilities. The majority of the changes can be done on the fly, without having to save the files temporarily on a hard-disk drive or a solid-state drive.6
Possibilities we actually have used so far include:
Change the container
ProRes is a very popular format not only for post-production, for which I assume it was originally designed, but also for capture. Therefore we, as archivists, have to find a way to preserve these native ProRes files, as well as the result of a ProRes-based post-production, because this is actually the highest quality available. QuickTime (.mov) is the «natural» container of the ProRes video codec. Unfortunately Apple has already limited the support of QuickTime format for Windows, and most probably they will limit its support on macOS soon too, with the launch of the fully 64-bit operating system. One piece of good news is that the bitstream syntax and decoding process of ProRes have been disclosed and published by SMPTE in 2015, therefore we do have officially some technical information about the format. Another piece of good news is that ProRes can also be wrapped into the Matroska container.
One change we sometimes do apply during data migrations is to transform files from ProRes encoded video that is wrapped in a QuickTime (.mov) container to ProRes encoded video in Matroska (.mkv) files. This is technically called trans-multiplexing (or transmuxing) and consists of the de-multiplexing (or demuxing) of the old files followed by a re-multiplexing (or remuxing) into the new files.7 This operation is also called re-wrapping.
The file can be trans-multiplexed (i.e. the file is first de-multiplexed and then re-multiplexed) very quickly, because transcoding (i.e. the extremely time-consuming decoding and re-encoding) of the file’s content is not required. This can be done easily, if needed, during a data migration, without any additional cost.
Change both the container and the video codec
During our own data migration we have transcoded and trans-multiplexed all the AVI files, containing HuffYUV or FFV1 version 0 or 1 encoded video, into Matroska files, encoded in FFV1 version 3 video for preservation purposes.
For different clients we often have trans-multiplexed and transcoded video content from stream-based Y′CBCR 4:2:2 uncompressed 10-bit video in AVI, QuickTime or MP4 files into lossless compressed FFV1 version 3 video wrapped into Matroska files.
Sometimes we have trans-multiplexed and transcoded from single-image-based DPX or TIFF 16-bit, 12-bit or 10-bit images into MXF, ZIP or TAR files (or even plain folders) into lossless compressed FFV1 version 3 video wrapped into Matroska or into lossless compressed JPEG 2000 wrapped into an MXF container, usually according to AG-07.
There is almost no limitation of the possible changes during data migration. The range spans from simply to perform actions, like modifying file naming conventions, to more time-consuming actions, like modify the metadata. This includes replacing the MD5 checksums by SHA-1 checksums, or fixing incorrect or missing information about frame rate, colour space, etc.
Data migration can also be used to delete unwanted files such as Apple’s «.DS_Store» and Window’s «Desktop.ini» files which may be written inadvertently onto the source cartridge.
In conclusion, taking account of our experiences so far, I am personally very confident that in the near future, when we will be obliged to migrate our company’s archive from LTO-6 to LTO-8, we will be able to replace the experimental containers and video codecs by more robust formats. This may be the case possibly in 2019 or 2020, after the launch of LTO-9, when the prices for LTO-8 will be diminishing.8
In my opinion, archives should adopt the formats that they can better master and handle today, knowing full well that their choices are not made forever. On the contrary, they will be able to change, if needed, during each of the coming data migrations at no or little additional cost.
I wish to acknowledge the help provided by Adrian Wood, Grover Crisp and Michael Friend.
Having graduated in mathematics and computer science, Reto Kromer became involved in audio-visual conservation and restoration more than thirty years ago. He was head of preservation at the Cinémathèque suisse (Swiss National Film Archive), and lecturer at the University of Lausanne and the Academy of Fine Arts, Vienna.
He has been running his own preservation company, AV Preservation by reto.ch, and lecturing at the Bern University of Applied Sciences. His current research includes colour spaces, look-up tables and codec programming and emulation.
This article has been published in IASA Journal, n. 49 (December 2018), IASA, p. 18–22.