In reply to a post by adebov:

I've done a bit more digging and the general consensus seems to be the number '10' is banded around as being a limit imposed by some web-servers (probably a DOS attack prevention measure). If such a server detects more than 10 concurrent connections from a single IP, they disconnect.

Well, interestingly the only case when I've seen what's obviously 50 or more download threads is bingbot indexing a site... and I checked that the IP address shows as "assigned ti Microsoft" in whois so it wasn't somebody spoofing the user-agent. (I noticed that because I got an alert from a server for what looked like a DOS, and given the source I'll call it MS-DOS - sorry).
So MS may enforce a limit in IE, but not in their own search engine.
(And I've reduced the limit on number of connections per IP, just in case they come back).

In reply to a post by Chrysalis:

its worth mentioning the defaults on threads in browsers such as firefox now breaches RFC, its impolite to use high numbers of threads to fetch content from a web server.

So the defaults on IE, FF and Chrome (which are all 6) are all in breach and Chrome is fixed (so you don't even have the option of reducing it). So all Chrome users are, by their very nature, impolite
BTW which RFCs? I assume you mean the http1.0 & http1.1 RFCs. They're big documents and I can't really be bothered searching through to find a one-liner concerning max connections initiated by the client (so I'm taking your word for it).

FF is higher than 6. its 20 last time I checked.

6 is ok without keepalive.

Edited by Chrysalis (Sat 19-Jan-13 21:48:53)

In reply to a post by adebov:

BTW which RFCs? I assume you mean the http1.0 & http1.1 RFCs. They're big documents and I can't really be bothered searching through to find a one-liner concerning max connections initiated by the client (so I'm taking your word for it).

Well, I read them from start to end once upon a time and so I could easily find this in RFC2616:
"Clients that use persistent connections SHOULD limit the number of simultaneous connections that they maintain to a given server. A single-user client SHOULD NOT maintain more than 2 connections with any server or proxy."
However these are more recommendation than mandatory limits (which would say MUST NOT - and BTW the ALL CAPS are from the RFC itself). Also RFC2616 has been updated by a number of more recent RFCs and I haven't checked if the above has been changed.
edit - the updates to RFC2616 don't seem to mention anything about max number of connections.

Edited by Uilebheist (Sat 19-Jan-13 21:53:27)

But if the problem is slow web-page loading when there are many links to pick up, which is where this question arose, surely those are not persistent links?

Not sure RFC2616 is very clear in this respect.
Just before the bit I quoted it states that all client must open connection assuming that the server will support persistent links, but may revise the assumption depending on the reply they receive. So it would follow that opening more than 2 links is frowned upon, unless you know in advance the server does not support keepalive. But they never say anything explicit.

I suppose the pertinent word is "maintain".
Is if frowned upon to temporarily open up more than 2 connections to aid speedy loading of complex pages (e.g. ones with dozens of images) provided those connections are closed once all elements are downloaded?
If 'keep alive' does, indeed, "keep alive" all opened connections (e.g. if you set 10 connections per server, so the web page completes faster, and 'keep alive' holds open the final 10 connections) then "we" have a problem - particularly Chrome, as the "max connections" is fixed at 6 (unless Chrome has a separate limit on keep alive connections).

The ambiguity is where RFC:2616 says "SHOULD limit the number of simultaneous connections that they maintain to a given server" but doesn't really define "maintain".
Is a maintained connection defined as one which is held open (e.g. if there are 10 max connections set up and there are 10 live streaming thumbnail videos, then presumably all ten are held open).
Is a connection held open only long enough for an image to be transferred (before closing and moving on to grab another element) not categorized as a "maintained" connection?

I'm going to quote figures from Firefox (only because they're easier to obtain than trawling through the registry to find the same figures for IE)...

network.http.max-connections default integer 256

network.http.max-persistent-connections-per-proxy default integer 8

network.http.max-persistent-connections-per-server default integer 6

So clearly FF default settings go against the limits defined in RFC:2616 (unless these are not counted towards the "maintained connections" - although the word "persistent" could be taken to mean "maintain").
In fact it seems only IE8 (and below) has a limit of 2 (and that's only for http1.0 - for http1.1 the IE default is 6, so that also goes against RFC:2616).

I find it hard to imagine MS, Google and Mozilla would collectively (and presumably; independently) decide to breach RFC:2616 unless the maintained (keep alive) connections are limited, elsewhere, to 2 - or unless those limits in RFC:2616 no longer apply (possibly because it was written 13 years ago, at a time when most office local LAN connections were slower than it's possible for home internet connections to go these days???).

In reply to a post by adebov:

I find it hard to imagine MS, Google and Mozilla would collectively (and presumably; independently) decide to breach RFC:2616 unless the maintained (keep alive) connections are limited, elsewhere, to 2 - or unless those limits in RFC:2616 no longer apply (possibly because it was written 13 years ago, at a time when most office local LAN connections were slower than it's possible for home internet connections to go these days???).

I wouldn't find that surprising. For example, there is a standard for cookies (RFC 2965), which none of these browsers seem to even know about (only Opera and a few other less-known browsers seem to comply!), the mechanism they use is from RFC 2109 which is not a standard and is marked as "obsolete". I could provide a login page on a site which is fully standard compliant, and only Opera users would be able to log in. Compliance for the "major" browser has always been a matter of "if it's convenient we'll do it".
As for the limit having been changed in the past 13 years, I checked the updates to RFC2616 (as listed in the RFC index I downloaded last night) and there is no change to that particular bit. Perhaps nobody felt it necessary to do anything since it is being ignored anyway.

In reply to a post by Uilebheist:

I wouldn't find that surprising. For example, there is a standard for cookies (RFC 2965), which none of these browsers seem to even know about

My guess is if Microsoft can't be bothered to make software that actually works properly, they're sure as Hell not going to bother reading the very wordy RFCs!

In reply to a post by Uilebheist:

I checked the updates to RFC2616 (as listed in the RFC index I downloaded last night) and there is no change to that particular bit. Perhaps nobody felt it necessary to do anything since it is being ignored anyway.

Quite possibly