Networking, storage and multiplatform production
Oct 5, 2007 3:00 PM
Network and storage resources can be divided into those that support production and those that are used for playout. In production scenarios, real-time presentation of the accessed media is not a performance requirement.
Content transfer and asset management system performance is dependent on file compression efficiency, disk access speed, network bandwidth, router latency and the number of terminal nodes (users). A centralized storage area network (SAN) will make content management easier in some respects, but the demands on the supporting media transfer network will increase because higher data transfer rates are necessary for optimal storage.
A network topology designed for storage access for non-real-time single-platform production may fail to function as needed for parallel production in support of multichannel distribution. Trying to remove bottlenecks caused by multiple concurrent storage access transactions by installing high performance, high accessibility storage is useless if the network doesn’t support sufficient data transfer speeds. The solution is in the engineering; each production workflow must be analyzed to develop a storage and network architecture that can support parallel production needs.
Perhaps a distributed storage system of production continents is a better solution than a centralized SAN. It all depends on workflow. Platform-specific network traffic can be limited to clearly defined domains, but total storage requirements will increase because each individual continent that supports each distribution channel will have its own storage system. This brute force method requires the management of numerous copies of content located in each continent.
OSI layer issues
A three-level network topology consisting of core, distribution and edge uses Open System Interconnect (OSI) layer 2 full wire speed, bandwidth-connected switching for the core and edge layers. With non-blocking layer 2 switches, all ports run at wire speed. So, a GigE interface truly runs at 1Gb/s. Even non-blocking switches, however, will introduce variable latencies depending on how many packets or frames are being buffered, so there is no guarantee about when packets will arrive at a node (networked device).
But things are different when packets reach the distribution level. The distribution level between the core and the edge is at OSI layer 3 and is connectionless and dynamic. This is where packet routing decisions take place and transfer duration speed indeterminacy is introduced. Multiple protocols, routing table updates and heartbeats that are acceptable on normal IT networks slow down packet transfers and are unacceptable on broadcast media networks.
Every active routing protocol will slow down network transfer rates, so the router must be locked down by turning off, or not activating, all unneeded protocols. An alternative is to use static routes. They require manual intervention to add more devices, but give the design team, rather than the device vendor, control of the packet routing.
Media IT
Media network design requires special consideration and alternative techniques. Generic IT methodologies that plug devices into any physically convenient switch or router and rely on VPN, VLAN and other network defining techniques can result in non-deterministic data transfer.
The use of static routing will facilitate the highest data rates, and, if properly designed, will deliver near 100 percent deterministic performance. Multiprotocol Label Switching (MPLS) circumvents router latency with techniques that enable layer 3 routing to approach layer 2 switching in performance. In addition, Quality of Service (QoS) packet tagging allows routers to prioritize packet dropping, if necessary. This helps get critical content where it is needed when it is needed.
Convergence is the updating of routers and switcher routing tables when there is a change in the network, and is dependent on routing protocols. Many standard routing protocols are virtually useless in media routing. For example, Routing Information Protocol (RIP) sends out router updates every 30 seconds, so content transfers may be fine, until the next routing table update. Then network congestion may cause packets to be dropped.
Even link-state advertising protocols, such as Interior Gateway Routing Protocol (IGRP), that only send updates when there are changes in the network, may disrupt packet transfers when consistent performance is needed most — in the event of device failure.
The addition of storage nodes will require routing table updates across the entire network topology for each active protocol. Some protocols can avoid this problem. Open Shortest Path First (OPSF) divides the network topology into hierarchical autonomous systems (AS). An AS is a group of routers that exchange information using link-state protocol. As resources are added to these zones, or domains, only routing tables in these areas must be updated, which speeds convergence.
| Want to use this article? Click here for options! |  |
