<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div class="moz-cite-prefix">On 10/26/2012 2:48 AM, Jay Daley wrote:<br>
</div>
<blockquote
cite="mid:B48B4CE8-5084-47C5-B52B-4CDCA5CCF089@nzrs.net.nz"
type="cite">
Either you are rate limiting at X qps in which case it should stop
in the first second where the traffic is < X or you are
actually describing the use of multiple buckets as I described
previously.</blockquote>
<br>
neither.<br>
<br>
<blockquote
cite="mid:B48B4CE8-5084-47C5-B52B-4CDCA5CCF089@nzrs.net.nz"
type="cite"> To put it simply, if you stop limiting 60 seconds
later then you are actually talking about X qpm not X qps.
</blockquote>
<br>
the log is 60 seconds late because that's how often idle buckets are
swept.<br>
<br>
but there's more to that answer.<br>
<br>
the window size describes the dimensions of a penalty box. the
victim's bucket gets some tokens every second, and it loses a token
for every attempted response. when the bucket contains zero or
negative tokens, responses are suppressed. but it can keep on going
more and more negative if more responses are suppressed than tokens
credited each second, up to a maximum of "rate * window". this
effectively means that if somebody is pounding the hell out of a
bucket they can remain in the "stopped" state for "window" seconds
after their behaviour improves. thus, a penalty box of "window"
dimension.<br>
<br>
all of this is explained at
<a class="moz-txt-link-rfc2396E" href="http://ss.vix.com/~vixie/isc-tn-2012-1.txt"><http://ss.vix.com/~vixie/isc-tn-2012-1.txt></a>. readability
patches are welcome.<cite><br>
<br>
paul<br>
</cite>paul<br>
</body>
</html>