[mythtv-users] Switching backend IDE drives to libata with 2.6.28 kernel

[Tom Dexter]

On Mon, Jun 1, 2009 at 2:23 PM, belcampo <belcampo at zonnet.nl> wrote:
> Tom Dexter wrote:
>>
>> I'm in the process of upgrading my various Gentoo machines from 2.6.27
>> to 2.6.28.  All the current recommendations call for moving from the
>> old IDE drivers (ATA/ATAPI/MFM/RLL) to the new SATA/PATA drivers
>> (libata).  I've made the move on one of my non-MythTV boxes and I have
>> some questions/concerns about the move, especially regarding my
>> backend machine.
>>
>> I'd especially like to hear from anyone using the libata with IDE
>> drives on their backend machines.
>>
>> A number of hdparm settings aren't supported with libata.  The idea is
>> to have the kernel just choose the ideal setting.  DMA gets set
>> automatically and isn't an issue.  32 bit IO (the -c option) isn't
>> supported, but all current opinion indicates that that setting is not
>> required these days.  The same goes for the interrupt-unmask (-u).
>>
>> The one that has be a little concerned is the multiple sector I/O
>> hdparm setting (-m) which isn't supported either.  From what I've seen
>> so far, libata seems to default this to 8 when the drives max is 16.
>> Except for WD drives, I don't see anything in the hdparm man page that
>> would lead me to believe that 8 would be better than 16.  As a matter
>> of fact, I never even used the -m option in hdparm, as the old driver
>> defaulted to 16.
>>
>> My backend has two 500GB Seagate IDE drives.  I haven't had a good
>> opportunity to get rid of my lvm partition in favor of storage groups.
>>  The lvm spans the drive that also has the / partition and the MySQL
>> databases.  I know that's not an ideal setup, but adding a drive would
>> be nightmarish, and in any case, it currently works fine the way it
>> is.  I can record three HD shows at once while watching another
>> previously recorded program with no problems at all.
>>
>> I really don't think I can afford to loose even a hint of disk
>> performance on that setup, which is why I started looking so carefully
>> at this one.  Any advise would be appreciated.
>
> 3 HD-shows at an unrealistic 33Mb/s calculates to 100Mb/s = 12.5MB/s
> Even an average drive from 2-3 years old does ->50MB/s, this is sequential I
> know, but never the less.

I agree that overall transfer rates wouldn't be an issue.  But I'm
still concerned about the multiple sector I/O setting because of all
the concurrent stuff going on including recording, playback, and
database reads/writes.  Even with my current setup, if I'm recording
three shows and also watching something at the frontend, starting up a
few commflagging jobs can cause an occasional IOBOUND error.  That
tells me I don't have much room to loose any performance.

Tom