Hello,

I plotted statistics of the latency (time difference between the market updates time given by IQFeed and the reception time on our servers).

I have 2 questions resulting from 2 problems I observed:

• It looks normal for all days except for May, 24th when something happened at around 9:30am (Paris time) as can be seen on this image: https://imgur.com/T5gHW4N . Latency went crazy, up to almost 4000 seconds of latency then it decreased linearly until coming back to normal. This may come from our servers but maybe you know about something that happened that day?
• Most plot look like this: https://imgur.com/vghWRAU . It looks like latency becomes higher after 6pm (Paris time). Why is that?

Thank you for your answers

Good morning,

I don't know that I am going to be much help here, but in evaluating the data I can see, I can say with relative certainty that we did not have a server side issue on the 24th that would have caused this issue. 4 seconds would be a huge issue for us, given that, I would have expected to find either other customer reports or internal communication about the problem and things look pretty dull on that day and given the timeframe, I don't see a way that we could troubleshoot this further.

As to the latency being worse in the evening, I have never heard of anything similar, so I am not sure what the answer there is either. It could be regular evening maintenance at providers between you and I which increases load to the switches between us, but we reserve all of our maintenance for on the weekends. Have you asked your ISP to see if they have insight? They may also have some tracking they could add. Beyond that, you could do a traceroute yourself at the time of failure to try and better understand where the latency is occurring.

Tim

Hello Tim,

Thank you for your answer.

Regarding May 24th, it may be a problem from our side, I need to investigate further. It's a complete outlier.

The server is in a data center in France, the provider (OVH) is the most important provider in France. I don't think there is any latency from our side.

If you look at the latency graphs for all days though, it looks like this latency happens on all days.

Please take a look: https://imgur.com/a/DchOn

I also added some box plots. Comparing to the 8am-6pm period which has a very decent latency with a median of 0.75 seconds, the 6pm-10pm period has a very high latency of almost 4 seconds.

Again, I don't think the datacenter the server is in could possibly increase the latency by 500% due to congestion!

Is something particular happening at 6pm everyday? I don't know what I am missing here :/

Just to clarify before we move on, you said 4000 seconds earlier, I am assuming that was intended to be termed to be milliseconds. When we look at the data you are showing here, is that supposed to be second data or millisecond data? Because a connection that was latent of any data for 2 minutes seems unlikely and so does a sub 10 ms across the world connection (at any time of the day). Now, what might make sense, is that this is all ms times and you are seeing connection speeds, with some degree of error being allowed for the system clock method of evaluation which cannot be perfect given the exchanges free will in syncing their servers to whoever they choose.

Also, the data you watch as of 6pm, how does that differ from what you watch during the day? Since our servers are located locally, is the data during the day only traveling from the US to you and then the data that is taking twice as long is having to travel from EU to US to EU and in turn increasing the latency?

Just things to consider.

Tim

I said 4000 seconds and I meant 4000 seconds haha! It surprised me too, it is really the only day I observed something like that. The data you sent us contained market updates from more than one hour (3600 seconds) earlier. Again, this is very strange and happened just on that day so we can just stop talking about it (except if that interests you, I have the logs of that day).

Since everything is in seconds, I do sometimes receive some level1 or level2 updates with dates from 2 minutes before. And after 6pm it becomes not so rare at all.

My program watching the symbols is not changing its behaviour after 6pm. It's keeping the connection open all day (all week) without changing its behaviour! So the data is always traveling from you (US) to us (EU).