Date: Thu, 4 Jun 1998 10:17:15 -0700 (PDT) From: Marc Slemko <marcs@znep.com> To: Karl Pielorz <kpielorz@tdx.co.uk>, Roger Marquis <marquis@roble.com>, Scot Elliott <scot@poptart.org>, Andreas Klemm <aklemm@hightek.com> Cc: freebsd-isp@FreeBSD.ORG Subject: Re: apache/freebsd limits on vhost servers Message-ID: <Pine.BSF.3.95.980604100912.5487A-100000@alive.znep.com> In-Reply-To: <19980603101623.05657@hightek.com>
next in thread | previous in thread | raw e-mail | index | archive | help
(replying to several messages here...) On Thu, 4 Jun 1998, Karl Pielorz wrote: [...] > Hmmm... 'Corrupt'? - We use them here, and we've never seen anything > corrupt... Nor have we had any problems with relative addresses etc. We make > sure that the default server for the IP has a nice page explaining how to > get to the domains if your browser doesn't support HTTP 1.1... I think it There seems to be a very common misconception that if a client doesn't send HTTP/1.1 requests and support HTTP/1.1, it won't send the Host: header. That is completely incorrect. While the Host: header is defined in the HTTP/1.1 spec, many HTTP/1.0 clients (eg. Navigator 4.x, IE 3.x) still send it. In fact, most of the clients sending it are HTTP/1.0 clients. [...] On Wed, 3 Jun 1998, Roger Marquis wrote: > On Thu, 4 Jun 1998, Leif Neland wrote: > > You can run all vhosts off the same ip; you don't need separate ip's. > > You can but it's not necessarily a good idea. All sorts of things can > corrupt a virtual server based on domain names only (as opposed to a > virtual server based on IP addresses) including various browsers, DNS > problems, dropped packets, relative references ... Erm... this is nothing but FUD. Yes, some clients don't support it. Other than that, please explain how DNS problems suddenly impact only name based virtual hosts. I have no idea why you think dropped packets will impact anything, and "relative references", whatever you mean by that, are not impacted at all. If you have some details, please give them. Otherwise, I have to say that there is little basis to your remarks other than the fact that not all clients support it. On Thu, 4 Jun 1998, Scot Elliott wrote: > That's not entirely true. > > Most modern browsers support it, but you'd be surprised at how many old > browsers are still being used. We put agent logging on our web servers > for a few months to test this; around 40% of clients did not have > browsers that could cope with host based virtual hosting (ie. based on the > Host: HTTP header rather than the IP address the request came in on). Your stats are very atypical. A far higher percentage of clients support the Host: header on nearly every website I have checked the stats. Are you sure you did not make a mistake in how you analyzed the logs? On Wed, 3 Jun 1998, Andreas Klemm wrote: > On Wed, May 27, 1998 at 07:38:05PM +0400, Alex G. Bulushev wrote: > > > At 09:17 AM 5/27/98 +0100, Karl Pielorz wrote: > > > >You should be OK with a couple of hundred - the real killer is when you get > > > >apache to write the log files for each virtual host to seperate files... > > > >This eats up file-descriptors... > > > > > > Is it better to log all virtuals into one single file? And if so what > > > analyzer works best with this setup? > > > > separate logs usefull for vhosts users, we use separate logs for 380 vhosts > > but FD_SETSIZE = 8192 > > What exactly influences this define ? The amount of filedescriptors > that can be managed per process ... I read something about a bitmap. > Would be nice if somebody could tell me. This define influences the size of the fd_set data structure used by select(). You can use descriptors over FD_SETSIZE just fine, presuming there aren't other problems that prevent it, however they will not work when you try to set them in a fd_set for use by select(). To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-isp" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.95.980604100912.5487A-100000>