On 20100701_001335, Stan Hoeppner wrote: > Anand Sivaram put forth on 6/30/2010 10:25 PM: > > > Why do you say that it is detected as IDE. Normally IDE disks using > > I don't get this either. Nothing in anything he posted shows that the kernel > is detecting this drive as IDE. Quite the contrary, it's being detected as a > SATA device, and if he'd have shown dmesg output, it would clearly state so, > but he did not. > > > deprecated IDE driver are shown as hda, hdb etc. where as SATA and the same > > IDE disks with newer PATA driver are shown as sda, sdb etc. For you it is > > showing the disk as sda. Take a look at "lspci -k" to see which kernel > > driver is getting used. > > Also a very easy method to see the reading speed of the disk is > > You're talking about libata, the current all-in-one SATA/PATA/ATAPI driver. > And yes, regardless of whether a drive is PATA or SATA, if it's under the > control of libata, it will show up as /dev/sdx, or if it's a CD/DVD-ROM as > /dev/srx. > > > dd if=/dev/sda of=/dev/null bs=1M count=1024 > > This will read the first 1024MB of your disk. I think a good > > disk/controller gives you more than 70MB per second or so. > > That depends on many factors, the big one being whether the drive and > controller both support NCQ, and if they both have a good implementation of > it. Look at the ATA_NCQ_HORKAGE list to see a group of drives whose > performance _drops_ considerably with NCQ enabled, or suffer other more > serious problems with NCQ enabled such as filesystem corruption, data loss, > etc.
I'm lurking here, hoping to learn useful stuff about hard drive software... What is NCQ? (in this context, of course) What is "ATA_NCQ_HORKAGE list"? The only hit that I get on this string in Google is a link to this email to which I am responding. TIA > > Other factors affecting sequential read performance (dd) are the elevator > used, and the nr_requests and read_ahead_kb settings. Bumping read_ahead_kb > up from the default 128 to 512 or 1024 will produce a decent increase in > sequential read performance, about 10-20%. For example, a quick test on one > of my lower end systems produces a 16% increase in sequential read > performance: > > /$ dd if=/dev/sda of=/dev/null bs=1M count=1024 > 1024+0 records in > 1024+0 records out > 1073741824 bytes (1.1 GB) copied, 16.8026 s, 63.9 MB/s > > /$ echo 1024 > read_ahead_kb > /$ dd if=/dev/sda of=/dev/null bs=1M count=1024 > 1024+0 records in > 1024+0 records out > 1073741824 bytes (1.1 GB) copied, 14.4375 s, 74.4 MB/s > > (This system only has only 384MB RAM so little to none of the performance > increase was due to buffers/cache from the first dd run) > > _But_ a high read_ahead_kb setting causes a huge rise in the size of kernel > I/O buffers, eating system memory like candy. This one test caused a 6 fold > increase in my kernel buffer size, to over 260MB. Playing with read_ahead_kb > for testing can be useful in measuring absolute hardware performance, but I > wouldn't run day-to-day with a setting much higher than the default. There > are some specific server applications where high read_ahead_kb is useful, such > as streaming media servers, but they are few and far between. > > -- > Stan > > > -- > To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org > with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org > Archive: http://lists.debian.org/4c2c23ff.7010...@hardwarefreak.com > -- Paul E Condon pecon...@mesanetworks.net -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20100701234757.gd21...@big.lan.gnu