Very slow writing to SATA disk

[Søren Schmidt]

On 29/10/2005, at 0:41, Mikhail Teterin wrote:

>> Look in smartmontools I provided patches for that, its not rocket
>> science you know...
>>
>
> This attitude -- on top of the API change itself -- is not really  
> encouraging
> for ISVs, you know :-)

Sigh, ataidle is a hack and the author had no intention to listen  
back when, so I dont feel teribly sorry about it you know. Spinning  
down disks needs to be done at the driver level so ATA knows what  
state the disk is in etc...
>
>> You need to find out what the transfer rates are for the RAW disk, ie
>> by doing a dd from /dev/zero to the disk with a a resonable blocksize
>> say 1M to minimize overhead. Also read speed from disk to /dev/null
>> blocksize 1M would be helpfull.
>>
>
> The read test averaged 62603828 bytes/sec over a minute or so.
> The write test is only 6931231 bytes/sec -- about 9 times less.
>
> While dd is running, `systat 1 -vm' reports about 110 ata- 
> interrupts (irq 25)
> per second during the write test, and about 1000 during the read test.

Hmm, that does sound as problems with that disk, or maybe disk vs  
diskcontroller. Any chance you could try the disk on something else ?

One other thing, how much mem do you have in there ? more than 4G and  
bounce buffering might get into the picture ruining the transfer rate...

Søren Schmidt
sos at FreeBSD.org