[189234] in North American Network Operators' Group
Re: NIST NTP servers
daemon@ATHENA.MIT.EDU (Lyndon Nerenberg)
Wed May 11 21:12:15 2016
X-Original-To: nanog@nanog.org
From: Lyndon Nerenberg <lyndon@orthanc.ca>
In-Reply-To: <20160511174234.A228E3F7@m0087796.ppops.net>
Date: Wed, 11 May 2016 18:12:03 -0700
To: surfer@mauigateway.com
Cc: nanog@nanog.org
Errors-To: nanog-bounces@nanog.org
> On May 11, 2016, at 5:42 PM, Scott Weeks <surfer@mauigateway.com> =
wrote:
>=20
> Wouldn't the buffers empty in a FIFO manner?
They will empty in whatever order the implementation decides to write =
them.
But what's more important is the order in which the incoming packets are =
presented to the syslogd process. If you're listening on TCP =
connections, the receive order is very much determined by the strategy =
the syslogd implementation uses to read from FDs with available data. =
I.e. elevator scan, lowest/highest first, circular queue, ... In a =
threaded implementation, your reader workers, buffer writers, etc., are =
all at the mercy of the threading implementation; it's difficult to =
control thread dispatch ordering at that level of granularity.
--lyndon
