Depends how close your subscribers are to the tickerplant. If you’re introducing network latency then you’re probably already too slow. Loopback on the sane host is very fast though (sub milli), so your main latency concern will probably be cpu. Make sure the tickerplant is tasksetted to a dedeicated core, with important subscribers on the same host.
Sent from my iPhone
the tickerplant process can be configured to push the messages as soon they arrive but i can buffer them to accumulate messages and reduce the network load with small packets.
so, it is matter of choice.
2012/12/24 Manish Patel <manni.patel@gmail.com>
Depends how close your subscribers are to the tickerplant. If you’re introducing network latency then you’re probably already too slow. Loopback on the sane host is very fast though (sub milli), so your main latency concern will probably be cpu. Make sure the tickerplant is tasksetted to a dedeicated core, with important subscribers on the same host.
Sent from my iPhone
There is indeed a sweet spot in latency. Smaller is not inevitablybetter. i do not think anyone has found it or have a good theory on ityet though. Our networks are well into that threshold and the variablecan be played with, even through a satellite link.greg~krsnadas.org–from: Felix LUNGU <felix.lungu>to: personal-kdbplus@googlegroups.comdate: 24 December 2012 00:12subject: Re: [personal kdb+] Ticker Plant / delay ?the tickerplant process can be configured to push the messages as soonthey arrive but i can buffer them to accumulate messages and reducethe network load with small packets.so, it is matter of choice.–2012/12/24 Manish Patel <manni.patel>Depends how close your subscribers are to the tickerplant. If you’reintroducing network latency then you’re probably already too slow.Loopback on the sane host is very fast though (sub milli), so yourmain latency concern will probably be cpu. Make sure the tickerplantis tasksetted to a dedeicated core, with important subscribers on thesame host.–On 23 Dec 2012, at 20:54, Aykut Uz wrote:Hi, I am looking into a document on KDB+ Tick created by FirstDerivatives(FD_kdb+tick_manual_1.0).I encountered the below: Ticker Plant:1. Receives the data from the feed handler. The ticker-plant storesthe data in memory for a shore period of time which is configurable.2. It logs updates to disk for recovery from failure and updates anysubscribers…6. The latency between the feed and the data being written to the logis less than 1 millisecond.Does that mean that any client subscribed to the Ticker plant willreceive updates with 1 milisend delay? That sounds not good, if youare going to do algo trading and if your algo engine is subscribed tothe ticker plant, this delay would cost you a lot.What do you think ?Aykut</manni.patel></felix.lungu>