Alternative to k3b? or why is everything soooooo slow?
Bruce Labitt
bruce.labitt at verizon.net
Tue Nov 8 21:40:01 EST 2005
Ben Scott wrote:
>On 11/6/05, Bruce Labitt <bruce.labitt at verizon.net> wrote:
>
>
>>Thanks for some insight on the problem. hdparm -i returns
>># hdparm -i /dev/hdc
>>
>>
>
> For finding out what hdparm "features" are set, the output of
>"hdparm /dev/hda" tends to be somewhat more useful, and certainly more
>concise.
>
>
>
>>It looks like it is udma4, doesn't it?
>>
>>
>
> That's what that appears to say. I'm not entirely sure how well
>that relates to the "-d" switch, though. IDE is screwy. For all I
>know, it could be in UDMA4 mode but not actually using DMA.
>
>
>
>>hdparm -c1 -d1 -u1 /dev/hdc means set IDE32 bit mode, set dma mode on,
>>and set unmaskirq flag?
>>
>>
>
> Yuppers. I find those three yield a great performance boost.
>Furthermore, I've never had any serious trouble with any of them on
>any hardware made in the past five years. The worst I've seen are
>systems that, upon the first access, immediately emit a message to the
>effect of "DMA didn't work, turning it back off". Never any data
>corruption or other nastiness. YMMV, of course, but I feel these
>three are pretty safe in this day and age.
>
> The link Ted posted also suggested "-c3", which I haven't been
>using. That might be worth trying.
>
> Said link also pointed out something I tend to just assume: *ALWAYS*
>back up your data. (I won't even say "...before trying anything."
>You should have backups. Period. If you don't, sooner or later, you
>will loose data. Guaranteed.)
>
>
>
>>What does set unmaskirq flag do?
>>
>>
>
> As Ted posted, it lets the system do other things while servicing
>the IDE bus at the same time. It results in a huge boost in
>interactive responsiveness, and should solve your clock lossage
>problem, at least.
>
>
>
>>Will the command string above change the dma mode?
>>
>>
>
> I'm not really sure. My guess is that "-X" sets the
>PIO/DMA/whatever mode, and "-d" tells the IDE driver to actually use
>DMA features. But that's just a guess.
>
>
>
>>It still thinks it is in udma4 mode, not PIO. Hmmm.
>>
>>
>
> PIO is "bad", so using UDMA is a good thing. One can ass-ume that
>the higher the number, the faster it is. :)
>
> As I understand it: PIO (programmed I/O) means the CPU has to handle
>feeding data to the IDE controller, in very small chunks. DMA (direct
>memory access) means the CPU just says "Hey controller -- transfer
>to/from this memory". That being said, when talking IDE, "UDMA" seems
>to sometime refer to the controller/drive interface, not the
>controller/RAM interface. I'm not sure how all that works.
>
>
>
>>As for media, I bought a bunch of RiDATA DVD+R's. I suppose I'll have
>>to pick up some other media to try, maybe some -R's...
>>
>>
>
> Like I said, check with the drive manufacturer for recommended
>media. It's not (just) a question of media quality -- it appears some
>drives just like certain media manufacturing methods/materials better.
>
>-- Ben "Backup often" Scott
>_______________________________________________
>gnhlug-discuss mailing list
>gnhlug-discuss at mail.gnhlug.org
>http://mail.gnhlug.org/mailman/listinfo/gnhlug-discuss
>
>
>
Thanks Ben. When I ran hdparm on /dev/hda [my hard drive] I got a
respectable
# hdparm -tT /dev/hda
/dev/hda:
Timing buffer-cache reads: 3616 MB in 2.00 seconds = 1808.90 MB/sec
Timing buffered disk reads: 174 MB in 3.00 seconds = 57.90 MB/sec
However, on /dev/hdc which is my optical drive I got only
# hdparm -tT /dev/hdc
/dev/hdc:
Timing buffer-cache reads: 3588 MB in 2.00 seconds = 1794.00 MB/sec
Timing buffered disk reads: 10 MB in 3.29 seconds = 3.04 MB/sec
This is pitiful, compared to the advertised 22MB/sec sustained read.
The above timing was performed after the following command:
# hdparm -X66 -c3 -u1 -d1 /dev/hdc
/dev/hdc:
setting 32-bit IO_support flag to 3
setting unmaskirq to 1 (on)
setting using_dma to 1 (on)
setting xfermode to 66 (UltraDMA mode2)
IO_support = 3 (32-bit w/sync)
unmaskirq = 1 (on)
using_dma = 1 (on)
Anymore ideas? I really don't know where to go from here. Thanks for
all the help so far!
Bruce
More information about the gnhlug-discuss
mailing list