FlashFire XP 32Bit (0.9f) - Another developing version

[zartoven]

This version... shows some information from its status window.
And... it can bypass big enough write requests.
This feature is for regular SSDs (not for netbook users)

When you install this version, you will be asked whether you are installing FlashFire for a netbook.
If you choose 'No', the write requests bigger than or equal to 64Kbytes will be bypassed to SSD.
This feature is beneficial for middle or hign-end SSDs even though the SSDs are already good enough.

As always, developing version may be unstable!

FlashFire09f.zip

(90.26 KiB) Downloaded 5308 times

[femp945]

Do you consider SSDs with the JMicron JMF602B controller "middle-end" drives that should bypass 64KB+ writes?

For example, OCZ Core V2, OCZ Solid, OCZ Apex...all use the JMF602B controller.

Also, is this version still only XP, or can it be used with Win7?

[Ennui]

Hi, would like to report that enabling and disabling works well even with a full buffer.

However, I vaguely recall reading that disc activity prevents disc flushing, is that right? I notice that when I'm playing music on Windows Media Player, the buffer usage starts to increase gradually, without flushing and sometimes reaches 100%. Only when I turn off the music, the buffer starts to empty. Is there any way to ignore light disc usage like music or videos?

If it's possible, perhaps the next release could allow us to define the range of write sizes to cache? For my SSD, the effects of caching becomes negligible after 4K width write stripes (as determined by ATTO), and thus the current 64Kb just causes too much to go into the buffer, such as while downloading files.

Thanks for this tool, it really is fantabulous =)

[zartto]

However, I vaguely recall reading that disc activity prevents disc flushing, is that right?

Right. Currently, FlashFIre does flushing when there is no disk activity including reading also.
(Maybe, I have to change the policy to flush when there is no 'write' request for a while for better reliability)

I notice that when I'm playing music on Windows Media Player, the buffer usage starts to increase gradually, without flushing and sometimes reaches 100%. Only when I turn off the music, the buffer starts to empty. Is there any way to ignore light disc usage like music or videos?

FlashFire does not allocate buffer for read requests.
So, if you played media files, the data will not occupy FlashFire buffer.
The reason, why the buffer becomes full, is ... in my guess, write requests from Windows OS.
For some reasons, Windows OS generates a lot of small, writes in background, and the writes are buffered by RAM buffer of FlashFire.
Because FlashFIre flushes its buffer when disk is idle, if you are playing media file, FlashFire thinks disk is busy, and the buffer becomes full.

If it's possible, perhaps the next release could allow us to define the range of write sizes to cache? For my SSD, the effects of caching becomes negligible after 4K width write stripes (as determined by ATTO), and thus the current 64Kb just causes too much to go into the buffer, such as while downloading files.

It is possible by modifying registry, and rebooting system.
However, it can be changed anytime.

HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\ffire\Parameters\WriteBypass

If the value is set to 0, all writes will be buffered.
If you want to bypass write quests bigger than 4K bytes, you can set the value with 9 (the unit is 512 byte-sized sector)

[The7]

This version... shows some information from its status window.
And... it can bypass big enough write requests.
This feature is for regular SSDs (not for netbook users)

Wonder whether you have tried it on your Kensington SSD?
If yes, would like to see the ATTO results for this version.

[zartto]

Wonder whether you have tried it on your Kensington SSD?
If yes, would like to see the ATTO results for this version.

Yes, sir

(1) without FlashFire

(2) with FlashFire without bypassing

(3) with FlashFire with bypassing

[Ennui]

If it's possible, perhaps the next release could allow us to define the range of write sizes to cache? For my SSD, the effects of caching becomes negligible after 4K width write stripes (as determined by ATTO), and thus the current 64Kb just causes too much to go into the buffer, such as while downloading files.

It is possible by modifying registry, and rebooting system.
However, it can be changed anytime.

HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\ffire\Parameters\WriteBypass

If the value is set to 0, all writes will be buffered.
If you want to bypass write quests bigger than 4K bytes, you can set the value with 9 (the unit is 512 byte-sized sector)

Thanks, this works well! (:

[mervincm]

I have used this with good results so far on the following 2 netbooks

ASUS EEEPC-900 2GB RAM (4GB FAST SSD, 8GB SLOW SSD, 8GB Class 6 SD Card)
Answered yes to the netbook question, and added the monitoring program to the start menu.
no other tweaks other than a 1024K offset and disabled indexing
Notebook is very usable performance wise, no identifed problems

ACER AOA-110 1GB RAM (8GB SLOW SSD, 8GB Class 6 SD card, empty SD card slot 2)
Answered yes to the netbook question, and added the monitoring program to the start menu.
no other tweaks other than a 1024K offset and disabled indexing
Notebook is very usable performance wise, no identifed problems

Looking forward to increasing the size of the cache, as I notice it fills easily when doing any patching or installing
Looking forward to increasing the amount of disks I can point it to, as both of my netbooks have 3 solid state storage

I am wondering it it would be worth setting a large bypass say 128KB or 512KB on even these slow SSD, as I think I might see more gain cacheing ALOT of small writes instead of a few large ones. I am talking real world benefit here, not artificial benchmark loads. Thoughts?
I am also wondering if there are any other tweaks that we might play with in the registry? In the past there were a few
Lastly I am wondering if flashfire is continuing to improve performance once the cache is full, or will performance drop to "non flashfire performance levels" till it manages to free up some space?

thanks again for the wonderful bit of code.

PS I am not able to run macrium reflex online imaging with Flashfire installed, I expect this is not something that can be worked around?

[The7]

I have used this with good results so far on the following 2 netbooks

ASUS EEEPC-900 2GB RAM (4GB FAST SSD, 8GB SLOW SSD, 8GB Class 6 SD Card)
Answered yes to the netbook question, and added the monitoring program to the start menu.
no other tweaks other than a 1024K offset and disabled indexing
Notebook is very usable performance wise, no identifed problems

ACER AOA-110 1GB RAM (8GB SLOW SSD, 8GB Class 6 SD card, empty SD card slot 2)
Answered yes to the netbook question, and added the monitoring program to the start menu.
no other tweaks other than a 1024K offset and disabled indexing
Notebook is very usable performance wise, no identifed problems

It seems that the bypass for low-speed random write SSD (most SSD netbook) would be good at 1024KB and that for medium-speed SSD (Kingston SSD) would be at 64KB. The bypass may be lower than 64KB for medium-high speed random write SSD.

One way to determine the bypass is to use the ATTO benchmark for the SSD without FlashFire. Use a value of KB near the flat top of the write-speed.


IMO. In SSD, the size of the erase page is 512KB for MLC. Buffer in write will be certainly helpful if the write size is less than 512KB. If the write size is larger than 512KB, the benefit of buffer in write will be decreased. It may slowdown the overall write speed for an very large write size.

[Ramon]

Hi Zartto,

FYI, the problem with the activation of the dirty bit and the consequent chkdsk at startup I reported in version 0.9d is still present in this version.

viewtopic.php?f=21&t=42&start=30#p261

Regards,

Ramon