ide write barriers
Quoted 2 lines
The goal is to make the use of write
back cache enabled ide drives safe with journalled file systems.
Does this mean that having write caching enabled is not safe if you are using ext3 on an IDE drive? Should "hdparm -W 0 /dev/hda" be used for example. (I see a 50% performance hit using "-W 0" when my box is under load.) If this is the case, what is the root cause? Do IDE drives reorder writes when they are cached?
Thank you in advance for any guidance.
Scott Lee
Quoted 8 lines · Scott Lee
Quoted 2 lines
The goal is to make the use of write
back cache enabled ide drives safe with journalled file systems.Does this mean that having write caching enabled is not safe if you are using ext3 on an IDE drive? Should "hdparm -W 0
/dev/hda" be used for example. (I see a 50% performance hit using "-W 0" when my box is under load.) If this is the case, what is the root cause? Do IDE drives reorder writes when they are cached?
As it stands, it's not safe to use write back caching on IDE drives. When the write completes as seen from the fs, it's not on the platter yet. That's a problem. And as you mention, there's no guarentee that writes won't be reordered as well.
So yes, either use the barrier patch or disable write caching.
Signature · 1 line
Jens Axboe
Quoted 8 lines · Scott Lee
Quoted 2 lines
The goal is to make the use of write
back cache enabled ide drives safe with journalled file systems.Does this mean that having write caching enabled is not safe if you are using ext3 on an IDE drive? Should "hdparm -W 0
/dev/hda" be used for example. (I see a 50% performance hit using "-W 0" when my box is under load.) If this is the case, what is the root cause? Do IDE drives reorder writes when they are cached?
All drives reorder unless instructed not to do so. ATA drives are just now getting FUA hooks. So flush cache often.
Cheers,
Andre Hedrick
LAD Storage Consulting Group