In the late 90s and early 2000s my company shipped out a Linux distribution on CD monthly, called KRUD. Every month we'd take the latest versions of packages for RedHat, plus some notable additions, and build updated media you could use to update or install Red Hat with. It was really nice to be able to install a new system, which we did a lot of, and have basically the latest packages, with Internet-based "yum update"s being fairly minimal after install.
So every month we'd burn, IIRC, nearly a thousand CDs and ship them out, mostly around the US. We were in a weird size where we had low enough volume and needed quick enough turn around, that going to big CD duplicators weren't really an option.
Anyone remember the Taiyo Yuden CD-R media? We spent a very long time trying various media and eventually settled on it as "the perfect CD-R". We spent years trouble-shooting problems that different people had, even paying people to ship defective media back to us so we could identify if it was burning problems or shipping problems. We had a shelving unit full of SFF PCs with CD-RW drives and some custom software I wrote that would burn, and verify the media.
Our shipped-defect rate went way down when I built some custom verification software. Initially I was just doing an "dd | md5sum" of the CD drive, after ejecting and re-loading the disc. But we eventually tracked user problems down to some drives having sporadic problems reading some sectors, which our drives would try several times and eventually get a good read of.
My new verification used low-level SCSI commands to read every block of data and get timing information, and if any sector took more than X ms to read, or more than Y sectors took more than X/5 ms to read, it would "fail" the disc. At this point we basically never had people reporting issues with the media.
Taiyo Yuden discs seemed to be the sweet spot between low cost and low defect rate.
Oh yeah, and there were all sorts of softwares to make the best of partial reads. I had one that I can't find the name of right now, that would compute forward-error-correction and make an additional file that you could store on the same disc, or on the last disc of a set. If you had 500MB of data, you could compute 150MB of FEC and fill the 650MB disc, and recover from up to 149.9MB of bad sectors later. (And that was the failures that got through the C1 and C2 error correction built into the drive, of course.)
There was another thing called jigdo, "jigsaw downloader", which would take a collection of files in a filesystem, and use them as the source to reconstruct a bit-perfect ISO. You could put the .jigdo file on the media itself, and the recipient could read the files, discarding the ones that failed, and (perhaps automatically?I don't remember this part) fetch only the failed ones over the internet, and reconstruct the original .ISO that you meant for them to have in the first place, with hopefully only a few minutes of dialup time.
Later, I found that BitTorrent could accomplish something similar. You could pre-seed your local copy of a torrent with burned media, flawed or perfect, didn't matter, just copy it to a writable hard drive. Launch your torrent client and point it to the "existing files on disk", and do a "force re-check", and it would go through and verify every block checksum according to the .torrent itself, joining the swarm to fetch the ones that failed and seed the ones that matched. In this way, if you had several discs with nonoverlapping errors, even without internet reachability, you could use local peer discovery make a LAN-only swarm and your clients would all share blocks and correct each other into all having a perfect copy.
I used this when I had a big pile of files that I wanted to hand out hundreds of copies of, to a hopefully-technically-savvy audience. Time was of the essence so I was burning at max speed and making my share of coasters, but I kept the failed burns, making sure that the .torrent itself on disc and the README describing this concept were readable, even if the rest of the burn was trash, and those were the ones I'd hand to people I knew could handle it. Sure enough, that evening I popped online and saw that a handful people had joined the swarm, most with a 99%-complete-already copy of the files, and they reached 100% in very short order, even the ones on dialup.
When I was buying them, in about 1995 (?), they ran about $20 a piece, IIRC. Every time there was a buffer under-run writing to one of them a little bit of me died inside.
So every month we'd burn, IIRC, nearly a thousand CDs and ship them out, mostly around the US. We were in a weird size where we had low enough volume and needed quick enough turn around, that going to big CD duplicators weren't really an option.
Anyone remember the Taiyo Yuden CD-R media? We spent a very long time trying various media and eventually settled on it as "the perfect CD-R". We spent years trouble-shooting problems that different people had, even paying people to ship defective media back to us so we could identify if it was burning problems or shipping problems. We had a shelving unit full of SFF PCs with CD-RW drives and some custom software I wrote that would burn, and verify the media.
Our shipped-defect rate went way down when I built some custom verification software. Initially I was just doing an "dd | md5sum" of the CD drive, after ejecting and re-loading the disc. But we eventually tracked user problems down to some drives having sporadic problems reading some sectors, which our drives would try several times and eventually get a good read of.
My new verification used low-level SCSI commands to read every block of data and get timing information, and if any sector took more than X ms to read, or more than Y sectors took more than X/5 ms to read, it would "fail" the disc. At this point we basically never had people reporting issues with the media.
Taiyo Yuden discs seemed to be the sweet spot between low cost and low defect rate.