Re: posix_fadvise()

[Guy Harris <guy () alum mit edu>]

On Apr 19, 2016, at 10:02 AM, Denis Ovsienko <denis () ovsienko info> wrote:

> As far as documentation goes, some sense may be presumed in libpcap calling posix_fadvise() with 
> POSIX_FADV_SEQUENTIAL. This way the OS would be able to adjust the buffers better to read the .pcap file. Does 
> anybody have any practical experience with posix_fadvise() to comment if the gain is ever visible with present day 
> hardware and OSes?

Annoyingly, on Windows it looks as if the equivalent is to specify FILE_FLAG_SEQUENTIAL_SCAN at *open* time, in 
CreateFile():

        https://msdn.microsoft.com/en-us/library/windows/desktop/aa363858(v=vs.85).aspx#caching_behavior

There doesn't appear to be an obvious way of turning that flag on after the file is opened.

It also appears that if you add the "S" flag to the list of flags in an fopen() call, you'll get "caching ... optimized 
for, but not restricted to, sequential access from disk.", presumably causing FILE_FLAG_SEQUENTIAL_SCAN to be set at 
open time, so that would be the equivalent on Windows:

        https://msdn.microsoft.com/en-us/library/yeby3zcb(v=vs.71).aspx

We already have to use a different flag to force binary interpretation of the file, so we might as well add "S" as well.

At some point we should add new APIs for reading capture files that works better with pcapng (but still supports pcap, 
so programs don't have to determine the file type themselves and choose which APIs to use); we should probably have 
those files offer random access (with "tell" and "seek" operations), and a way to indicate whether the file is being 
opened for sequential or random access.
_______________________________________________
tcpdump-workers mailing list
tcpdump-workers () lists tcpdump org
https://lists.sandelman.ca/mailman/listinfo/tcpdump-workers