Disks:

The disk geometry is very complex.
The abstraction hardware developers offer is that nearby
blocks are "close" w/r/t to access time.

The time spent on a disk access is:
1) Queue time
2) Overhead
	-- seek
	-- rotational delay
3) Access
	Size of access / Transfer rate


Efficiency of disk = (transfer time) / (positioning time + transfer time)

50% efficiency is a reasonable goal.

-----------------------------------------------------------

Scheduling on disks:

First-Come First-Served (FCFS)
-- pro: simple, fair
-- con: lots of scanning

Shortest Seek-Time First
-- pro: simple, good disk overhead usage (though suboptimal)
-- con: unfair, starvation potential

Elevator Algorithms (SCAN). The head sweeps left to right and back, 
servicing requests if they are in the pass.
pro: somewhat fair (except for edge-effects), decent performance

------------------------------------------------------------

Solid State Drives (SSD's):

random read: 25 ns
seq read: 30 ns
page program: 300 ns
block erase: 2 ms

Over-writes: very expensive.

Much faster on reads, but same ballpark with the block-erase for writes.

SSD's are much lower power. Useful for mobile devices.
Better schock resistance.

Lower capacity / $.

Flash cells wear out. 

Flash translation layer is analogous to a log-structured file system (coming soon).

------------------------------------------------------------

Other options (research):

memristors, ???

------------------------------------------------------------

File Systems:

A file system is a persistent data structure built on top of the disk abstraction.

Persistence: data is in a consistent state at all times, data lasts even if
the device fails (power outage).

Persistent data should not depend on temporary data, like virtual addresses.

Interface:

Create a a file

Delete a file

Read <file,offset>

Write <file,offset>

Misc. tasks


Some truisms:

-- most accesses are reads
-- most programs access files sequentially & completely
-- most files are small, but large files take up the majority of space

File systems stores file attributes (aka metadeta) about each file. eg:
1) file size
2) last access time
3) permission bits
4) owner
5) logical block number (lbn)

file-name is NOT part of the metadata. It is part of the directory.

Root of a file is called (in unix) the index node, or inode.
Contains metadata, info about how to access the file.

---------------------------------------------------------------

Moral equivalents to virtual memory:

base-and-bound <--> contiguous allocation
-- inode just stores lbn.
Pros: simple, sequential access wins
Cons: fragmentation is horrible.

Segmentation for vm <--> Extent.
Cons: same as segmentation.

Paging for vm <--> each inode contatins a translation table.
Pro: no fragmentation
Cons: cost of accessing random blocks is expensive,
	size of inode varies.

multi-leevl page table <--> multi-level indexed file.