Joe Orton wrote:
Can someone on Solaris have a look at this bug and check whether the sendfilev64 use is OK?
First I can already tell you that as of two months ago, Sun had acknowledged a bug that sendfile's performance was subpar to standard write/writev one copy mechanics in userland space. [See Colm's blog on T2000 testing for his observed performance shift attempting to use sendfile.] I can only guess there are other 'interesting' artifacts when you start talking about external network mounts. The sendfile pragma is almost never worth using and potentially harmful in all of these cases. Opening this bug with Sun would make the most sense at this point. Bill
