On Thu, Oct 10, 2013 at 6:00 PM, Greg Freemyer <greg.freemyer@gmail.com>wrote:
Mandeep Sandhu <mandeepsandhu.chd@gmail.com> wrote:
On Thu, Oct 10, 2013 at 5:00 PM, Ulka Vaze <ulka.vaze@l2it.com> wrote:
Hi, disk cache is same as filesystem cache. Also called buffer cache. This is implemneted below fs layer. It is basically a cache of disk blocks mainatined in RAM. (In pages) called buffers.
Ok. So this won't contain "files" but rather "blocks" many of which will represent a single file?
The buffer cache knows nothing about files. It may happen to contain the blocks that correspond to a complete file, but the buffer cache doesn't have any way to know that.
The purpose of this cache is to improve performance as disk devices are slow. You can access this cache from the kernel. Block layer accesses this from the request structure and commits blocks on disk. There are more layers in between like IOschedulers / SCSI etc.
Where does the mapping for file to disk pages/blocks exist? Is it in the inode or dentry entries or something else?
That is a very filesystem dependent question.
In general dentry entries point to inodes and inodes point to blocks of pointers. Those pointers point to actual data blocks.
You need to discuss a specific filesystem type to even start to discuss anything specific.
Ie. Does the fat filesystem have an equivalent of inodes?
How does your device accesses files ?
The device itself runs stripped down version of a fairly recent Linux version (3.x). It has DMA capabilities to transfer content to/from the hosts memory from/to it's own.
If 3.x is really current you have 2 more options at least: dm-cache and bcache.
I don't know much about either, but they are going to be better options for I think than the traditional buffer cache. Greg -- Sent from my Android phone with K-9 Mail. Please excuse my brevity.
Ulka, Greg, Thanks a lot for your inputs. I have a lot of reading up to do, so I'll come back if have any specific doubts. Thanks again. Regards, -mandeep