[linux-cifs-client] cifs version 1.26 test version for 2.6.8 or later kernels

[Gorazd Golob]

Hi!

I'm testing reading big (~100mb) files from w2k share. Here is
/proc/meminfo. No services are running on computer.

MemTotal:      1035196 kB
MemFree:          5020 kB
Buffers:          1480 kB
Cached:         998732 kB
SwapCached:          0 kB
Active:          27668 kB
Inactive:       979820 kB
HighTotal:      131008 kB
HighFree:          140 kB
LowTotal:       904188 kB
LowFree:          4880 kB
SwapTotal:     1959920 kB
SwapFree:      1959920 kB
Dirty:          104024 kB
Writeback:           0 kB
Mapped:          10000 kB
Slab:            17168 kB
CommitLimit:   2477516 kB
Committed_AS:    23324 kB
PageTables:        592 kB
VmallocTotal:   114680 kB
VmallocUsed:      1216 kB
VmallocChunk:   113328 kB
HugePages_Total:     0
HugePages_Free:      0
Hugepagesize:     4096 kB




On Wed, 1 Dec 2004, Steven French wrote:

> My worry would be if there are many scenarios in which large file copy
> causes huge spike in cache memory (cifs inode data cached in page cache),
> then competing memory requests cause write back of page cache which gets
> blocked on the memory request which is in turn blocked on the write back
> of the page cache.  Linux can be very aggressive in caching.
>
> I was able to reproduce some interesting problems when repeating VERY
> large writes (perhaps 100MB or more for each write) in parallel to
> multiple files to samba server on the same box (which probably won't
> reliably work in loopback without changes to writepage itself or
> cifs_writepage or alternatively moving part of smbd to kernel)
>
>
> Steve French
> Senior Software Engineer
> Linux Technology Center - IBM Austin
> phone: 512-838-2294
> email: sfrench at-sign us dot ibm dot com