First post / slow burns - 3520 on linux

Hi, all, I’ve done a bunch of searching, most of it has lead to, but I’ve still come short of finding an answer to my problem.

I recently bought a silver NEC 3520A from I installed the drive in my Debian linux box, kernel 2.4.27, and was able to burn my first data disk right away. But right away took about 30 minutes for 2 GB. I have tried a few more burns, and they all start out at 12x and end up averaging right around 2x. Forcing the burn to 8x has the same results.

Media: Verbatim 8x DVD-R MCC 02RG20.

Software: growisofs by <>, version 5.21, front-ending to mkisofs: mkisofs 2.01-unofficial-iconv (i686-pc-linux-gnu)

Firmware: stock 1.04 - I’m holding off on an upgrade until I determine the drive doesn’t have to go back to newegg.

IDE: I’ve had the drive as slave and master and even put it on its own Promise Ultra100 TX2 IDE controller. It was actually slightly slower on it’s own.

hdparm reports that the drive is in UDMA2. Oddly, hdparm also reports that the drive has no buffer, though I believe it should be 2MB. I also can not set MultSect size, a common speedup for newer drives. The hard drives in the system are set to UDMA5.

Tonight, I’m going to try to patch a Windows box together to do some testing, but so far I’m fairly disappointed with my situation. The 0kb buffer thing has me wondering if this is a bad drive, but perhaps hdparm just isn’t understanding it. So far I’ve made 3 data discs and 2 video discs. At some point, the first I burned became a coaster - it MD5SUMmed after I made it, but now craps out though there are no physical defects. The two video DVDs play well in my player. So it burns, it just burns slowly.

The last burn:
1.6G /mythtv/dvd
real 9m58.075s
user 0m0.110s
sys 0m3.230s

((1 632 MB) / (598.07500 seconds)) / (1 385 (KB / second)) = 2.01750532x

I’m looking for 1) other linux users, since I didn’t see evidence of many on the board 2) other linux users with a 3520 - what does hdparm tell you? 3) suggestions on other media to try 4) suggestions on the 0Kb buffer thing 5) any other advice you can offer up.

Sorry if this rambles, it was a lot clearer in my head, but I was up too late last night monkeying with this thing.

Thanks for any help you can offer.

As a followup, I did a fresh install of Win2K on the same hardware. I’m now burning with Roxio Easy Media Creator Basic DVD (came with drive) and 15% into a 2 GB burn, I’m averaging 1.4x (I set it to burn @ 8x).

I didn’t mention before, but this is an AMD Athlon XP 2500+ on an Asus A7N8X-X w/ 512MB RAM. Both drives are Masters (HD on pri and DVD on sec.) and in UDMA mode.


I burn at 4x on my linux box with 4x media. I couldn’t get hold of a better media yet. I use k3b for burning and I set hdparm parameters like this :

hdparm -q -X udma2 -d1 -c1 -u1 /dev/hdd

Also I use 2.6.11 kernel which doesn’t use scsi interface for cd/dvd burning so it should be faster.

Thanks for the response.

That’s what I’ve been feeding hdparm, too. When I do a timing test this is what I get:

hdparm -Tt /dev/dvd

Timing cached reads: 1524 MB in 2.00 seconds = 762.00 MB/sec
Timing buffered disk reads: 26 MB in 3.06 seconds = 8.50 MB/sec

That 8.5 seems slow to me, but that’s not really based on anything. What does your drive report? I think tonight I’ll try the 2.6 kernel. I’ve got some tricky hardware, so I don’t go into a kernel upgrade lightly, but I’ll give it a shot. BTW, I don’t have ide-scsi set up, growisofs doesn’t need it, only for burning CDs. At least that’s what I understand. Maybe that’s part of my problem.

My wife will be so pleaseed to know I have another night with the computer planned. . . .

Timing cached reads: 520 MB in 2.01 seconds = 258.87 MB/sec
Timing buffered disk reads: 26 MB in 3.19 seconds = 8.15 MB/sec

That looks normal.

i think the ata controler is read as ascci,i use kb3 but in that box i have an old 811(lite-on) and it only will go 4 but it is not used a lot

Cartman, thanks for the reference. As a follow-up, I compiled the kernel yesterday and was able to burn 4.2GB @ 7.3x average and reportedly hit 10x a few times. So I now have more confidence in the drive itself. I can’t full-on upgrade to 2.6 yet as I’m running LVM and that needs to be upgraded to LVM2, which is a one-way street. I’m going to do some more testing and research before I go down that path as there’s over 200GB of TV shows and about 50GB of mp3s at risk if I botch LVM (this is a MythTV box).

Thanks for the help guys. The conclusion seems to be that if you want speed with this drive, burn under kernel 2.6.11.

Cool :slight_smile:

Also you might want to enabled “Packet CD/DVD Writing” in kernel but that gave some unusable CDs here with k3b so I don’t personally recommend it but its supposed to make cd/dvd burning faster.

I do not have a 3520 just yet, nor will I be dropping back to a 2.4 kernel. For what it is worth, using my 3500 under 2.6.9 FC3, I have been a lot happier with cdrecord from cdrtools-2.01a32-ossdvd.patch than growisofs from dvd+rw-tools. You will likely have to compile this yourself though and have a bit of fun with the devices. On a 2.6 kernel running as root I enter:

cdrecord dev=ATA -scanbus

… to determine device and then burn with:

cdrecord dev=ATA:1,0,0 driveropts=burnproof -dao gracetime=2 speed=6 my-iso-file.iso

If you are combining the creation of the iso and the burning in one command, this might not work so well. I generally have an iso handy.

If I do not drop the speed=6, then it will use the default burn strategy of 8-12x and I prefer the scan of a 6x burn over the default strategy. When the speed goes up on the default strategy towards the end, so does the graph of the scan. I go for quality over speed, and 6x is not that much slower than 8x. Specifiying speed=8 will still use default strategy. I have only been using TYG02 lately.

Might be trying NERO for Linux soon.

Just to clear up a few things stated here :slight_smile:
I’ve been using linux for many many years, and here are a few clarifications that will help:

  1. If I remember correctly, using linux ide-scsi (which shows ATA devices as /dev/sg0 etc…) does NOT use DMA, but PIO… This can account for slowness in this situation - be SURE to be using a newer 2.6 kernel that allows SCSI passthru from the ATA devname.

  2. hdparm MultSect is only for disks, not optical drives… I don’t know where you read this is a ‘obvious speedup’, but it only works and effects PIO rates even, doesn’t do anything for DMA access. (please see from the Con Kolivas patchset mailing list, a post from Jens Axboe, who is a I/O hacker for the linux kernel.)

  3. hdparm is for what it says, mostly, hard drives… Running a hdparm test on a optical drive will give incorrect results.

  4. Even though Schilling (cdrecord developer-not too good on people skills though) may hate you for it, you can address cdrecord’s dev= with a device name (eg. dev=/dev/hdd) … It works fine, ignore the warning. Please though, refrain from using cdrecord or one of its hacked varients for burning DVD’s under linux, it is a horrible hack for DVD writing and will not give good results in many situations, and also ruin or corrupt your ability to burn CD’s! Please see growisofs ( If you have problems with growisofs, I’d look into them more and solve them.

  5. Packet CD/DVD writing is completely seperate from burning cd’s with things like growisofs and the likes… It shouldn’t make any difference in normal burning situations, and from what I hear it is still quite buggy.


I’ve managed a 16X burn on my Benq and Liteon drives using Suse 9.2 and KDE.
There is an option to set UMDA 33 or UDMA 66 modes in this distro for optical drives. It works very well for me on a fairly modest system.