From Bugzilla Helper: User-Agent: Mozilla/5.7 [en] (Win98; U) Description of problem: My new laptop has a CD on /dev/hdb. It is a 1.6GHz but performed much slower than the old 0.5GHz laptop. After some testing i became convinced that the settings in cd-capplet to check data or audio CDs interferes rather badly with linux if the controller is shared with a disk that contains the system. The task at fault is called magicdev. Version-Release number of selected component (if applicable): How reproducible: Always Steps to Reproduce: 1.check data or audio CD action 2.hdparm -t /dev/hda a few times 3.uncheck it 4. hdparam -t /dev/hda a few times Actual Results: 2-> very fluctuating , from 8 to 18 MB/sec 4-> consistent 21 MB/sec Expected Results: the checking should perhaps be done less often, or perhaps could check if the CD is on IDE and shared with / or so.... in that case perhaps lower the frequency. Or perhaps add an interfce to cd-capplet to allow the user to select a frequency of checking (how is this done, just polling??? there was nothing obvious to magicdev) Additional info: The fact that i didnt notice this on the old laptop is probably because they don't share the IDE controller (old one used /dev/hdc). I tried measuring it on my desktop, but the CD is SCSI and found it (not too surprisingly) not to matter too much. It's a serious error though, the buttons below don't let me choose an appropriate option it seems since there isn't loss of data or memory, just unacceptably slow :-)
I just checked a new fresh installation of rh73 on the same model laptop (a Dell 8200) and by default the settings of cd-capplet are such that the IO problem will be present. On the good side of things, KDE seems to suffer from the same problem. A casual check of the sourcecode of magicdev appears to show devices are checked only every 1000 ms, or 1s. So, it's still a mystery to me how it can interfere with traffic on /dev/hda so badly. I have another machine where I believe i also have a CD shared with /, this would be another testcase. It could also be that the hardware on this laptop is peculiar as to aggrevate this 'problem'. Btw, most of the Dell8k series laptops should have this problem, since they most have a CD or DVD on /dev/hdb. Some have a a DVD in hdb, and the CD in hdc, if two multimedia are used. I decided to get the combo version instead.
In general, I very much that sharing a bus should be sufficient to trigger this problem; more likely its a bug in your CD-ROM drive. I don't see any possible interval that would be both non-intrusive and useful; it sounds like all that's really possible for you is to turn off the automount feature. Can you provide the model information from /proc/ide/hd*/model ? A blacklist in magicdev is possible.
Also, do you get any output in /var/log/messages?
more /proc/ide/hd?/model :::::::::::::: /proc/ide/hda/model :::::::::::::: IC25T060ATCS05-0 :::::::::::::: /proc/ide/hdb/model :::::::::::::: HL-DT-STCD-RW/DVD-ROM GCC-4240N and there are not suspicious - or any for that matter - messages in /var/log/messages. Btw, note again that KDE seems to suffer from same. We also tracked down a solution, i.e. Programs -> Settings -> Peripherals -> CD Properties. It also brings up a window very similar to cd_capplet, and if turned off the IO is acceptable again, and hdparm will show a normal throughput on an idle system (which has been our testcase to test good vs. bad).
I have a Dell C840 with the exact same harddrive and CD model combo. I was observing very slow startup times and reported it in bug 73661, but only determined the problem was with 'magicdev' recently. How do I disable magicdev in gnome? Currently I just kill the process. Also, I installed RH 8.1 beta (Phobe) and in that release magicdev doesn't seem to exhibit the same problem.
*** Bug 83260 has been marked as a duplicate of this bug. ***
*** Bug 89718 has been marked as a duplicate of this bug. ***
I'd like to add that I have a Dell Inspiron 8200 with a DVD/CDRW drive and magicdev was causing excruciatingly slow performance, which I only discovered after a lot of searching on the web. The only solution I've discovered is rpm - e magicdev. I've read other people having this problem (a reviewer noted that his Open Office launch times were 90+ seconds --- I was getting 100 seconds on my 2 GHz Dell), so it might not be restricted to Inspirons.
*** Bug 90149 has been marked as a duplicate of this bug. ***
One thing that would be interesting to know is whether the commands that magicdev is sending to the drive are taking a long time to complete. You should be able to determine this by, when logged into GNOME. A) Make sure that magicdev is enabled in the 'gnome-cd-properties' dialog. B) Run: $ killall magicdev $ strace -T -o /tmp/log -e open,close,ioctl magicdev Wait for about a minute then kill it. C) Attach the resulting log file to this bug report
Created attachment 91578 [details] Log file output for magicdev
The above strace output that I just attached was just while running magicdev on an Inspiron 8200 with no other activity. I can also try running an strace while trying to do something else (launch OpenOffice)... as noted above, while running magicdev on an I8200, launching OpenOffice takes 100+ seconds, as opposed to 15 seconds, normally.
*** Bug 97134 has been marked as a duplicate of this bug. ***
If you have the Gnome CD-Player panel application running, the performance problem exists even if magicdev is not running.
Closing bug. Magicdev has been removed in favor of gnome-volume-manager in devel branch.