FancyCache suggestion high performance and high latency sug

[Skeilio]

1st off FancyCache is already a wonderful application that satisfies most of my needs Thanks.
When will pricing be available?

A couple of things I think would be nice to have are some write optimization methods

If deferred writing is enabled
An item that has 2 options "Read Around Write", and "Read Cut-Thru Write"
1st option is where if there are drive writes and reads at the same time FancyCache
keeps writing while reading
2nd option is where there is writing if a read occurs writing is paused for # ms 1-10000
this allows reads to always have priority on the spindle.
This should be an option as Raid Controllers can easily handle simultaneous read-write
where a single spindle would suffer due to seek times.

I have also noticed on some sata ssd , usb, iscsi sans, and aoe sans that write defered writing can be much slower
than having write deferred disabled.
Lets take a file copy ~600MB like a iso in which is already has been read cached into ram either via fancy cache or the OS
The target copy device is a drive mounted via iscsi via DroboElite, NetGear ReadyNAS, MS Target 3.3, or Starwinds product
If write Deferred is disabled and you copy the file to the drive you can see write performance around 30MB to 85MB a sec
depending on product.
If write deferred is turned on write performance never typically gets above 9MB sec so I am assuming the block size being used to write
is to small.

Have you thought about read ahead cache? some programs do not read there data files using large block sizes
but instead read anything from a few bytes at a time to a few KB
This can help on files that have indexes or look-up tables and can prevent unneeded additional drive seeks or additional small reads
on files like certain flat file databases or game data files this is especially apparent on disks with high rotational or command latencies.

If you sell this product for server systems you may want to have some additional options for the level 2 cache
If the max read-write times becomes greater than a certain threshold level 2 cache would automatically be disabled
also when reading data from l2 cache implement as an option to have a crc check to insure data has not been damaged
as failing flash and ssds can return bad information.

Also for server and multiprocessor systems you may want to look at NUMA Non-Uniform Memory Access.

OK its time to stop now dont you hate feature creep?

Thanks

[FrozenLord]

Great idea
I would love to have "Read Cut-Thru Write" - I am on a cheap SSD and each write-request kills / slows down a parallel read-request...
(This results in an almost freeze while Fancy Cache is flushing its cache, because all read-requests get significantly slowed down)

So I think that a Read Cut-Through Write would help quite a lot in my scenario.

[Support]

Hi Skeilio,

Thank you very much for such great suggestions!