On Fri, Jan 28, 2011 at 9:33 PM, Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> wrote: > Hello, > > On Friday, January 28, 2011 4:31 PM Alan Stern wrote: > >> On Fri, 28 Jan 2011, Marek Szyprowski wrote: >> >> > This commit add code that flushed page cache disk buffers every 128 read >> > and write request. It has almost no impact on the performance of the >> > emulate usb drive but saves system page cache from being filled with >> > useless emulated disk data. Could you also pls confirm the impact of this change on the performance of the mass storage device? >> > >> > Signed-off-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> >> > Signed-off-by: Kyungmin Park <kyungmin.park@xxxxxxxxxxx> >> > >> > --- >> > >> > Hello, >> > >> > In the embedded system development we encoutered a really anoying >> > behaviour of USB mass storage gadget. We use it to export SD card as USB >> > disk. However when user copies a lot of data from or to the USB disk, >> > the mass storage gadget consumes almost all pages from system page >> > cache. This results in a significant degradation of system and io >> > performance. A lot of valid and really often used disk cache data is >> > wiped from memory and need to be constantly reloaded. >> > >> > Caching mass storage backing file is usually useless. Host system very >> > rarely sends requests for the same blocks in the limited time window. >> > Usually there is another caching layer on the host system anyway. >> > >> > Disabling page cache for the backing file is also not the solution, >> > because mmc perform really bad on direct 512 byte reads and writes. >> > >> > I've did a quick hack to limit the amount of page cache that is being >> > used by mass storage gadget. I've forced mass storage gadget to flush >> > and invalidate page cache once every 128 read or write request. I know >> > that the patch is ugly but it solves the issue. >> >> Can you post an equivalent patch for file_storage.c? The same issues >> apply there. > > Right. I wonder if my approach is really correct. Flushing every 128th > request seems to be really hacky. I would really like to find a better > way to decide when to flush/invalidate the page cache. > > I will post a patch for file_storage.c on Monday, now I need to go out > from the office. > >> > Do you have any suggestion how to correctly introduce a method of >> > limiting page cache usage in the mass storage gadget? >> >> How do other storage or filesystem servers do it (e.g., NFS, CIFS)? > > I will investigate how this is handled there. Thanks for the comments. > > Best regards > -- > Marek Szyprowski > Samsung Poland R&D Center > > -- > To unsubscribe from this list: send the line "unsubscribe linux-usb" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
