Post-production storage: how Hot, Archive and Deep tiers fit a show

Most post-production storage is still bought one show at a time: a RAID or a NAS big enough to hold every card from every shoot day, kept on fast disks until delivery. That is simple, and for a short project it works. On a series it means paying for fast storage on many terabytes of footage that nobody opens between the edit and the online.

A show doesn’t use all of its media at the same time. This post goes through what each stage actually reads, how much bandwidth that takes, and how CloudStudio’s three storage tiers (Hot, Archive and Deep) line up with it.

What each stage of a show reads

Ingest and dailies. Camera originals are written once, checked, and used to make proxies. After that, the originals are mostly not read again until the online.

Offline editorial. Editors work from proxies. A DNxHR LB proxy runs at about 36 Mbps at 1080p24, roughly 4.5 MB/s per stream and about 16 GB per hour of footage. ProRes 422 Proxy is about 45 Mbps, roughly 20 GB per hour. Even a team of editors scrubbing several streams each needs modest bandwidth, but they all need low latency and the whole proxy library online.

Online and color. The finishing team needs the camera originals, but only for the shots in the locked cut. These are heavy. Apple lists ProRes 4444 at about 330 Mbps at 1080p29.97 in its ProRes white paper, and UHD needs several times that. In our own testing, 4K ProRes 4444 needs about 125 MB/s per stream to play without stuttering. Camera RAW is often higher.

After delivery. The show is done, but contracts often require keeping the camera originals and masters for years. This media is rarely read, if ever.

So the media that needs to be fast changes over the life of the show. During offline it is the proxies. During the online it is a small slice of the originals. After delivery it is almost nothing.

The three tiers on CloudStudio

Hot. Hot is fast shared storage for the media people are working on this week: proxies during offline, and restored camera originals during the online and color. It mounts as a normal network drive on CloudStudio workstations, so editors and colorists see it as folders. Our storage nodes connect at 100 GbE, and our colorists grade 4K ProRes 4444 straight from Hot.

Archive. Archive holds camera originals between the edit and the online. It costs much less to keep media here than on Hot, and it is not meant to be edited from directly. When a folder moves from Hot to Archive, every file is checked against its Hot copy before the move counts as done. In our own moves we have measured 446 to 686 MB/s from Hot to Archive, which is around 1.6 to 2.4 TB an hour.

Some vendors call this middle tier nearline. On CloudStudio, Archive covers that role.

Deep. Deep is tape-backed storage for media a show has to keep after delivery but doesn’t expect to use. It is the cheapest tier. Restores from Deep take longer and are arranged on request, so it is the wrong place for anything a cut might still need.

RAID or NAS per show versus tiered storage

A per-show RAID keeps everything on the same fast disks for the whole schedule. Capacity has to be bought for the largest the show will ever get, and the drives have to be physically where the editors are, or shipped. Getting a second copy offsite is a separate job.

Tiered storage keeps only the working set on Hot. Camera originals sit on Archive for most of the schedule, and Hot grows and shrinks with what people are actually using. Editors and colorists on CloudStudio workstations all mount the same storage, so there are no drives to ship.

The trade-off is the restore. With a RAID, the originals are always there. With tiers, they have to come back before the online, so the restore has to be fast and has to bring back the right files.

How much of a show actually needs to be on Hot

Less than most people expect. Two numbers from real shows:

A DNxHR LB proxy hour is about 16 GB. A 4K ProRes 4444 hour is about 450 GB at 125 MB/s. The proxies for a show are around 3 to 4 percent of the size of 4K ProRes 4444 originals.
For the online of one docuseries episode, the conform pull brought back 505 camera files totalling 3.88 TB. The show archive was 115.6 TB, so the online needed 3.4 percent of it.

The rest of the show can stay on Archive until the next cut needs something from it.

Getting media back for the online

Most archive storage is slow at the step that matters: getting things back. Cloud archive classes often take hours before the first byte is available, and then somebody still has to work out which files the cut uses.

On CloudStudio, the post team uploads the XML from the locked cut and the conform pull restores only the camera originals that the timeline uses, back to Hot, with every file checked against its archive copy. On the docuseries above, that took 2 hours 34 minutes for 3.88 TB. Read the full job numbers.

Choosing a tier

Media people are working on this week, including proxies and the originals for an online: Hot
Camera originals between offline and online: Archive
Delivered shows you are required to keep: Deep

See how a show moves through all three, or read how edit proxies are made during upload.