Download Webcasting: Dealing with significant audiences behind the

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
March 2010
Webcasting: Dealing with significant
audiences behind the corporate firewall
Ed Van Petten
CIO / Vice President,
Network Operations
ON24, Inc.
© 2010 ON24, Inc. All Rights Reserved.
Introduction
Webcasts sometimes involve significant numbers a single corporation’s employees in supporting the events
(manning booths and the like) and/or visiting the events as guests. In some cases, the Webcast is intended to
accommodate a corporate meeting – such as a sales meeting, where virtually all the attendees are from a single
corporation, many attending from a network inside the corporate firewall. In these situations, it is essential to
assess the demands that such events place upon the corporate network, and to provide methods to minimize any
negative impacts – on both the networks and the Webcast.
1
© 2010 ON24, Inc. All Rights Reserved.
Loads: What you should be worried about
When a Webcast occurs, there are several loads:
• The Interface – mostly graphic elements and code – that presents the show experience
• Chat traffic , if applicable
• Webcasts - Most webcasts are comprised of audio or video streams along with slides and various modes of
interactivity. The media streams all arise from servers outside the corporate firewall.
The first two of these loads are not significantly different than the normal browser traffic that a corporate network
and its gateways are designed to handle routinely.
Chat traffic is supported by Jabber servers, which use XMPP as their protocol for communication. ON24’s
implementation tunnels the XMPP traffic using BOSH, which is a polled protocol using HTTP
(http://xmpp.org/extensions/xep-0124.html). The practical effect of using BOSH is to have each client make
continuing HTTP request every 5 seconds. The XMPP traffic is then passed in both directions in the “1 per 5
second” package. There is no traffic other than port 80 traffic, and the packets individually are small, consistent
with chat activity.
The webcast traffic, however, can have characteristics that will load the network gateways significantly – perhaps
to the extent of congesting the gateway and interfering with other traffic.
The traffic of interest is that which is simultaneous (i.e., a large number of participants initiate a stream all at once)
and demanding (i.e., audio streams at 20K are not especially demanding, while video streams at 300K are more
demanding).
Webcast Types:
• On Demand Webcasts – usually elements of content available for selection by the show participants at their
option. These events lack the characteristic of simultaneity. While they may have 300K video streams, they
are simply not accessed by large numbers of people at the same times.
• Live webcasts – these are frequently keynotes or other addresses of significant interest. The speakers may
be well known and capable of drawing a crowd. This is almost always a subject of concern if the media is
video.
• Simulive webcasts – these are webcasts which are actually pre-recorded but presented as though they were
live. The attendees cannot distinguish that the presentations are not live. These events share the
characteristics of live events in terms of the potential for network load.
So, it’s clear that loads to be concerned about include live and video – audio streams are lightweight enough to
ignore and on-demand webcasts do not generate the simultaneity required to cause a problem. There is another
significant area of concern – VPN access. It is common to have corporate VPN access set up such that the
default route for all the network traffic points to the corporate VPN gateway. Even web traffic that is entirely
unrelated to the corporate network flows through the corporate VPN gateway inbound – and then out the
corporate gateway to the internet, returning by the same path. This is in contrast to the arrangement where only
traffic destined for the corporate network gets routed to the VPN, while other traffic goes directly to the local ISP.
If all traffic flows through the corporate VPN, then media traffic may overwhelm the VPN gateway, even if the
corporate gateway is capable of supporting the indicated volume of traffic.
It also seems obvious that a corporate network where all the traffic flows through one or a few internet gateways
is vulnerable in a different way than a network which is organized with each site having its own gateway to the
internet. Of course the size and utilization of the gateway determines the scope of the vulnerability
2
© 2010 ON24, Inc. All Rights Reserved.
Sensitivity Analysis
It should be clear, then, that the IT staff needs to address the following points:
• With respect to the live and simulive webcasts,
• in Video format,
• what is the expected attendance
• per internet gateway location
• and/or for VPN users who route media streams through the VPN gateway
3
© 2010 ON24, Inc. All Rights Reserved.
Measures to Address Vulnerable Networks
The analysis suggested above should yield a list of sites which are vulnerable given the expected audience. If
one or more sites are problematic, there are a options available to address the situation. They boil down to
limiting the number of internal stations playing the event or re-distributing the stream inside the network to
minimize stress on the gateway.
Option 1: Limiting the number of internal stations
•
•
•
•
have participants gather in a conference room, providing one projected stream rather than one per viewer.
Block access to the stream through the corporate firewall – which protects the network but prevents access to
the stream
Password protect access to the event, and limit the distribution of the valid password
Limit the distribution of the access URL to a small group of personnel
Option 2: Redistributing the stream inside the network
Plainly, it is possible to locate a source for the stream inside the corporate network. Several options exist:
•
•
•
Multicasting - In the event that the network is multicast enabled, and the multicast is reliably known to be
functional in all locations, it is possible to use it to distribute the streams inside the network. Using
multicasting, a stream is fetched from the head end media server and replicated to a multicast-capable media
server internal to the customer network. This customer multicast server actually delivers the stream to the end
users.
Unicast Replication - It is also possible to configure unicast replication. In this scenario, Windows Media
servers are deployed in the client network, and publishing points set up to enable to replication of streams
from the head end server. End users connect to the local streaming servers, fetching the streams from a local
network location rather than through the internet gateway. This relieves the capacity strain on the inbound
gateway. Unicast Replication and Multicasting are further documented in the Advanced Signal Services PDF,
which is available upon request.
Peer-to-peer stream propagation – Multicasting typically requires a considerable lead time to get active and
tested in all locations; unicast replication requires the deployment and maintenance of physical servers. Many
companies elect to deploy software which receives a live stream and redistributes the stream in the local
network. This peer-to-peer distribution strategy works well and avoids the issues which stand in the way of the
multicasting and unicast replication strategies. It does rest on a client which downloads to the client desktop,
which presents an obstacle in some networks. ON24 partners with Octoshape for its peer-to-peer
implementation; further information is available upon request.
Launch Mechanics
When an event is launched on the ON24 system, the launch sequence (see below) results in a playlist being
provided to the end users; this playlist provides access to the media stream. In the situation described here,
where different clients need to access the media through different servers or technologies, logic has to be
provided which inspects the user network location and generates a playlist specific to that situation. For example,
• External users – get sourced from the external CDNs
• Internal Users –if the client is from 10.2.1.0/24, select and the Dallas media server, for example. Playlists and
the criteria to select them are manually built by ON24 Signal Acquisition.
4
© 2010 ON24, Inc. All Rights Reserved.
Usual Launch Sequence
Lobby Page
1
Client Audience Console
Audience URL
Launch
Media
Player
3
2
Slides
4
Decide which
playlist to deploy
based on “network
location” of client
Registration Page
(Optional, and
bypassed after first
access, if used)
Playlist provides list of
media assets to be
accessed in sequence
Usually, it is not possible or worth the effort to address every possible network location; there are generally a list
of sites where you’d like the media play attempt to fail, and the client access to be provided by a non-network
constrained approach. ON24 provides events that support media streams for the AV distribution and
simultaneously allow the audio to be accessed by means of conference calls (non-streaming events). It is usual
to configure 80-90% or the corporation’s users to access via streaming, but to support the small or remote
network locations by provisioning non-streaming capability providing conference call access to the signal.
Importance of Failover Mechanics
The playlist controls what happens when a server can’t be accessed by a particular user. Consider these two
scenarios:
Assume we’ve got media being distributed through a set of unicast replicas to an audience in Cincinnati.
5
© 2010 ON24, Inc. All Rights Reserved.
1
X
am
am
re
Stre
St Media Server
1X
Media Server
New York
Dallas
Stream Fetch X 1
1 X Stream
1
X
1X
a
re
St
Internet Gateway
Head End Media Server
and Firewall
Los Angeles
m
Stre
am
Media Server
Cincinnati
Stream Fetch
Strea
m Fetc
h
Str
ea
m
Fe
tch
Media Server
Little Rock
Media Server
Miami
Scenario 1. Cincinnati Server Working Properly
In this scenario, the bandwidths to the unicast replica servers are all at 1 times the stream speed. This is the
typical set up. Consider the impact if the Cincinnati media server goes down, and the playlist New York listed
below Cincinnati.
New York to Cinci
Link is loaded at 3X
am
re
St Media
eam
Str
Server
1X
Dallas
1 X Stream
Media Server
New York
m
tch
h
Fe
etc
mF
am
re Strea
St
1
X
St
re
a
0 X Stream
Fe
tc
h
1
X
a
re
St
Internet Gateway
Head End Media Server
and Firewall
Los Angeles
1X
m
Stre
am
Media Server
Cincinnati
Media Server
Little Rock
Media Server
Miami
Scenario 2. Cincinnati Server Out of Service
It seems clear that selecting a sequence of servers in the playlist which avoids the risk of overloading a link in
case of server failure is important. ON24 will assist in identifying and evaluating failover scenarios.
VPN behavior presents an additional risk. Generally, the VPN infrastructure is comprised of one or more VPN
servers which are supported by gateway capacity (bandwidth facing the internet-located remote users). This
gateway capacity may be limited. Also, since the VPN gateways are encrypting and shipping the bits from the
local network resources to the remote users, handling media streams can place demands on the devices which
they are not configured to support.
VPN configurations generally provide routing information to their remote clients which differentiates between
traffic to be sent to the corporate network and traffic to be sent to the ISP supporting the remote user. In Cisco
terms, this is referred to as “split tunneling” = some traffic is sent to the VPN tunnel (i.e., the corporate site) and
other is split out of the tunnel and handled locally. Many corporations elect to route all traffic through the VPN
tunnel, including the media server traffic for webcasts. This concentration of traffic through the VPN facilities and
associated gateways presents a risk of congestion.
A simple resolution is to route the media server traffic directly to external media servers. If this is unacceptable,
the risk can be mitigated to some extent by locating a media server adjacent to the VPN servers.
Complex failover behavior can be configured to reflect the various network scenarios in place.
6
© 2010 ON24, Inc. All Rights Reserved.