![]() ![]() No consumer ethernet technology can currently guarantee link recovery in less than 50 ms. If the user really cannot afford to lose an ethernet frame, and there is no buffering-retransmission on the Dante endpoints, you need two separate networks. with 2 cables), in case of fault of one of the switch uplinks, loss of ethernet frames will happen both using LACP and STP, with spanning tree being far worse. In case of redundant connection between switches (i.e. Tbh for non-professional users, I would rather invest in a cheap unmanaged switch with a 10 GbE uplink rather than fiddle with QoS in a low-end consumer switch, whose support and latency impact depends again on the switch's firmware. A Dante 192 Khz/24bit stereo flow is only ~50 Mbps IIRC, so again for domestic use a normal 1 Gb ethernet would be OK, unless there is a download-heavy teenager (or a NAS) in the network path between the two Dante endpoints. Latency spikes should not happen, unless the switch's chipset is total garbage or there is routing in the middle (and again, with a poorly underscaled router). For professional use in a large studio, totally worth it. After all, these network variables (packet loss, latency, and jitter if one is curious) are measurable.Īgree, VLANs would only protect from broadcast/multicast (depending on how the switch OS handles it), which is a non-issue in a small domestic network. My main point was to not discount interface "displacement" in favor of features - if a switch has both, and enabling QoS does not cause other issues (= it does not increase the switch CPU cycles, which can happen if the switch's dataplane chipset sucks), why not. ![]() I've never used a Dante system, so i don't know how sensitive it is to loss of a single UDP datagram. Some QoS implementations work really well, others (especially in consumer gear) still can suffer from drops under some conditions. However, if the incast congestion is so bad to be a problem, QoS might be a palliative vs just a larger link. In these situations, I agree that QoS will help as long as it is honored down the path. That all said, this is getting very nitty-gritty, and in practice on a reasonable switch, following basic guides with default settings you're unlikely to run into an issue.ĭante is UDP though, as far as I can tell, so kind of immune from the window throttling behavior you correctly explain.Įxcessive incast problems on a PHY will result in latency only if the switch has non-shallow buffers, otherwise the result will be discards. In comparison, AVB switches handle this on an even lower layer, by reserving "Time Domains" with a scheduler that can guarantee packets flow in an exact expected timeframe, achieving microsecond consistency, making it useful for SDI video systems that need genlock as well. the process of initially starting an upload (lets say starts an upload test) the computer will often spike the PHY rate in several spots, before the speed settles and the packets coming into the switch become more evenly spaced out.ĭSCP flags resolve this problem almost entirely, and is one of the reasons they recommend a switch with such feature. ![]() ![]() So even if you have internet with only say 20mbps upload, and your LAN is gigabit. This is never an issue for PC users of course but a Dante system might care if you're targeting very low latencies. Even if the connection isn't quite maxed, just the process of scaling TCP window sizes around involves hitting many types of choke points on a network. If there's no QoS, this can lead to latency spikes (on the order of 500 - 1000μs) on a shared switch link further down the chain. Many non-realtime OSes (Windows, macOS) tend to buffer and vomit full size frames onto the network, in bursts. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |