How many concurrent AJAX (XmlHttpRequest) requests are allowed in popular browsers?

369
votes

In Firefox 3, the answer is 6 per domain: as soon as a 7th XmlHttpRequest (on any tab) to the same domain is fired, it is queued until one of the other 6 finish.

What are the numbers for the other major browsers?

Also, are there ways around these limits without having my users modify their browser settings? For example, are there limits to the number of jsonp requests (which use script tag injection rather than an XmlHttpRequest object)?

Background: My users can make XmlHttpRequests from a web page to the server, asking the server to run ssh commands on remote hosts. If the remote hosts are down, the ssh command takes a few minutes to fail, eventually preventing my users from performing any further commands.

8
ajaxbrowserxmlhttprequest
Thinking about your situation, what is the feasibility of pinging the remote hose to see if it is up or down? This won't answer your question, but this may be a better workflow. – Bob
Thanks Bob, that's one of the two approaches I had planned to fix this problem -- I considered mentioning it in the Question but decided it was off-topic. (Another approach is to have the server, which I control, timeout the ssh requests.) – Michael Gundlach
I think you pretty much have your answer... it's more than safe to assume Safari and Chrome support at least 2, so you can always assume 2. – Rex M
Using Chrome 2.0.172.28 on Windows Vista I got 6 concurrent connections. – Callum
I just found this page stevesouders.com/blog/2008/03/20/… which gives a few more numbers and a discussion about this. – David Johnstone

8 Answers

145
votes

One trick you can use to increase the number of concurrent connections is to host your images from a different sub domain. These will be treated as separate requests, each domain is what will be limited to the concurrent maximum.

IE6, IE7 - have a limit of two. IE8 is 6 if you have a broadband - 2 (if it's a dial up).

102
votes

The network results at Browserscope will give you both Connections per Hostname and Max Connections for popular browsers. The data is gathered by running tests on users "in the wild," so it will stay up to date.

25
votes

With IE6 / IE7 one can tweak the number of concurrent requests in the registry. Here's how to set it to four each.

[HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\Internet Settings]
"MaxConnectionsPerServer"=dword:00000004
"MaxConnectionsPer1_0Server"=dword:00000004
7
votes

I just checked with www.browserscope.org and with IE9 and Chrome 24 you can have 6 concurrent connections to a single domain, and up to 17 to multiple ones.

6
votes

According to IE 9 – What’s Changed? on the HttpWatch blog, IE9 still has a 2 connection limit when over VPN.

Using a VPN Still Clobbers IE 9 Performance

We previously reported about the scaling back of the maximum number of concurrent connections in IE 8 when your PC uses a VPN connection. This happened even if the browser traffic didn’t go over that connection.

Unfortunately, IE 9 is affected by VPN connections in the same way:

6
votes

I have writen a single file AJAX tester. Enjoy it!!! Just because I have had problems with my hosting provider

<?php /*

Author:   Luis Siquot
Purpose:  Check ajax performance and errors
License:  GPL
site5:    Please don't drop json requests (nor delay)!!!!

*/

$r = (int)$_GET['r'];
$w = (int)$_GET['w'];
if($r) { 
   sleep($w);
   echo json_encode($_GET);
   die ();
}  //else
?><head>
<script type="text/javascript" src="http://ajax.googleapis.com/ajax/libs/jquery/1.4.2/jquery.min.js"></script>
<script type="text/javascript">

var _settimer;
var _timer;
var _waiting;

$(function(){
  clearTable();
  $('#boton').bind('click', donow);
})

function donow(){
  var w;
  var estim = 0;
  _waiting = $('#total')[0].value * 1;
  clearTable();
  for(var r=1;r<=_waiting;r++){
       w = Math.floor(Math.random()*6)+2;
       estim += w;
       dodebug({r:r, w:w});
       $.ajax({url: '<?php echo $_SERVER['SCRIPT_NAME']; ?>',
               data:    {r:r, w:w},
               dataType: 'json',   // 'html', 
               type: 'GET',
               success: function(CBdata, status) {
                  CBdebug(CBdata);
               }
       });
  }
  doStat(estim);
  timer(estim+10);
}

function doStat(what){
    $('#stat').replaceWith(
       '<table border="0" id="stat"><tr><td>Request Time Sum=<th>'+what+
       '<td>&nbsp;&nbsp;/2=<th>'+Math.ceil(what/2)+
       '<td>&nbsp;&nbsp;/3=<th>'+Math.ceil(what/3)+
       '<td>&nbsp;&nbsp;/4=<th>'+Math.ceil(what/4)+
       '<td>&nbsp;&nbsp;/6=<th>'+Math.ceil(what/6)+
       '<td>&nbsp;&nbsp;/8=<th>'+Math.ceil(what/8)+
       '<td> &nbsp; (seconds)</table>'
    );
}

function timer(what){
  if(what)         {_timer = 0; _settimer = what;}
  if(_waiting==0)  {
    $('#showTimer')[0].innerHTML = 'completed in <b>' + _timer + ' seconds</b> (aprox)';
    return ;
  }
  if(_timer<_settimer){
     $('#showTimer')[0].innerHTML = _timer;
     setTimeout("timer()",1000);
     _timer++;
     return;
  }
  $('#showTimer')[0].innerHTML = '<b>don\'t wait any more!!!</b>';
}


function CBdebug(what){
    _waiting--;
    $('#req'+what.r)[0].innerHTML = 'x';
}


function dodebug(what){
    var tt = '<tr><td>' + what.r + '<td>' + what.w + '<td id=req' + what.r + '>&nbsp;'
    $('#debug').append(tt);
}


function clearTable(){
    $('#debug').replaceWith('<table border="1" id="debug"><tr><td>Request #<td>Wait Time<td>Done</table>');
}


</script>
</head>
<body>
<center>
<input type="button" value="start" id="boton">
<input type="text" value="80" id="total" size="2"> concurrent json requests
<table id="stat"><tr><td>&nbsp;</table>
Elapsed Time: <span id="showTimer"></span>
<table id="debug"></table>
</center>
</body>

Edit:
r means row and w waiting time.
When you initially press start button 80 (or any other number) of concurrent ajax request are launched by javascript, but as is known they are spooled by the browser. Also they are requested to the server in parallel (limited to certain number, this is the fact of this question). Here the requests are solved server side with a random delay (established by w). At start time all the time needed to solve all ajax calls is calculated. When test is finished, you can see if it took half, took third, took a quarter, etc of the total time, deducting which was the parallelism on the calls to the server. This is not strict, nor precise, but is nice to see in real time how ajaxs calls are completed (seeing the incoming cross). And is a very simple self contained script to show ajax basics.
Of course, this assumes, that server side is not introducing any extra limit.
Preferably use in conjunction with firebug net panel (or your browser's equivalent)

4
votes

Wrote my own test. tested the code on stackoverflow, works fine tells me that chrome/FF can do 6

var change = 0;
var simultanius = 0;
var que = 20; // number of tests

Array(que).join(0).split(0).forEach(function(a,i){
    var xhr = new XMLHttpRequest;
    xhr.open("GET", "/?"+i); // cacheBust
    xhr.onreadystatechange = function() {
        if(xhr.readyState == 2){
            change++;
            simultanius = Math.max(simultanius, change);
        }
        if(xhr.readyState == 4){
            change--;
            que--;
            if(!que){
                console.log(simultanius);
            }
        }
    };
    xhr.send();
});

it works for most websites that can trigger readystate change event at different times. (aka: flushing)

I notice on my node.js server that i had to output at least 1025 bytes to trigger the event/flush. otherwise the events would just trigger all three state at once when the request is complete so here is my backend:

var app = require('express')();

app.get("/", function(req,res) {
    res.write(Array(1025).join("a"));
    setTimeout(function() {
        res.end("a");
    },500);
});

app.listen(80);

Update

I notice that You can now have up to 2x request if you are using both xhr and fetch api at the same time

var change = 0;
var simultanius = 0;
var que = 30; // number of tests

Array(que).join(0).split(0).forEach(function(a,i){
    fetch("/?b"+i).then(r => {
        change++;
        simultanius = Math.max(simultanius, change);
        return r.text()
    }).then(r => {
        change--;
        que--;
        if(!que){
            console.log(simultanius);
        }
    });
});

Array(que).join(0).split(0).forEach(function(a,i){
    var xhr = new XMLHttpRequest;
    xhr.open("GET", "/?a"+i); // cacheBust
    xhr.onreadystatechange = function() {
        if(xhr.readyState == 2){
            change++;
            simultanius = Math.max(simultanius, change);
        }
        if(xhr.readyState == 4){
            change--;
            que--;
            if(!que){
                document.body.innerHTML = simultanius;
            }
        }
    };
    xhr.send();
});
0
votes

I believe there is a maximum number of concurrent http requests that browsers will make to the same domain, which is in the order of 4-8 requests depending on the user's settings and browser.

You could set up your requests to go to different domains, which may or may not be feasible. The Yahoo guys did a lot of research in this area, which you can read about (here). Remember that every new domain you add also requires a DNS lookup. The YSlow guys recommend between 2 and 4 domains to achieve a good compromise between parallel requests and DNS lookups, although this is focusing on the page's loading time, not subsequent AJAX requests.

Can I ask why you want to make so many requests? There is good reasons for the browsers limiting the number of requests to the same domain. You will be better off bundling requests if possible.