dd command

[jok llamera]

hi fellas,

When I write directly to a character device, does it bypasses buffer cache? if it is, then does it used async device? how?

fig.1 # dd if=/var/adm/sa/sa01 of=/dev//r bs=8192

Another thing, when I write directly to a filesystem using dd command does it use buffer cache or not?

thanks,
joks

[Tom Danzig]

I believe dd write directly to the device regardless of whether or not you use the block or character device file. I'm not 100% sure though.

FYI, in addition to "fellas", there a quite a few very sharp ladies in this forum!

[Sridhar Bhaskarla]

Yes. It doesn't bypass the buffer cache. All raw device disk transactions will go directly to the disk subsystem not via buffer cache.

So your dd command will bypass the buffer cache.

Whenver a file system is accessed it goes through buffer cache unless the vxfs option of mincache=direct is specified.

-Sri

[Sridhar Bhaskarla]

sorry.. a small correction in the top like.

Yes. It does bypass the buffer cache...

-Sri

[A. Clay Stephenson · Accepted solution]

Hi:

If you write to ANY character device /dev/rdsk/c2t2d0, dev/vg01/rlvol1 there is no buffer cache. If you write to ANY block device /dev/dsk/c2t2d0, /dev/vg01/lvol1 the buffer cache is used. Buffer cache is always used in filesystem writes unless you enable the OnlineJFS vxfs options convosync=direct,mincache=direct which bypass the buffer cache. This applies to any application not just dd.

Clay

[Anthony deRito]

It is good to use sar -b to analyze cache hit ratios. (Or Glance if you have it) I find these utilities allow you to understand more clearly whats going on with disk I/O. Raw I/O never uses buffer cache... the lower overhead makes it faster. Whenever you shut down a system or run sync manually, all
modified buffers of the buffer cache are flushed (written) out to
disk.

Tony