If you think about it, every system is (broadly understood: often the users or even developers are part of the system) taking in facts about the world, doing some sort of analysis and then taking action.
This is a great talk. Thanks, Kevin. At 27:23, are the Backup OR's literally on standby? That is, are they not processing requests at all? If so, what impact have you seen from transitioning to the Backup OR's? If not, do you ever run out of memory on a single machine? Furthermore is state shared across machines in this case?
Nice job. My 2 cents: If you're not measuring wire latency, and you are committing a lot of complexity to measure microsecond latency, than really your RTT latency is in the milliseconds, probably 5-10 milliseconds if you are cohosted at the exchange. Better focus on measuring and optimizing millisecond latency and ignore the microseconds until you have some spare time and budget.
In todays trading infrastructure, there is nothing that takes millisecond, if you are that slow you lost. Trading on FPGA are measured in tens of ns, algos right now hover around 1 microsecond.
@@dota2tournamentss The weird irony is that different market participants coexist that are working at typical latencies that vary by 10,000 to 1. The fastest three sell side equity shops are much slower than the best buy side , yet also 100 times faster than small regional banks.
@@kevgol0 hi thank you so much for your response I am a recent graduate who just wants to learn about low latency java development, I haven't found much online other than things similar to your video and quickfix j manual
This is lunacy. How can you talk about HFT and distributed computing in the same sentence? The latency of communication between two neighboring computers is cosmic compared to the numbers this guy is talking about (20-25 microseconds). And in the first part of his talk this guy talks about kappa-architecture (!) and data lake storage represented by a database (!!!). That's what people from analytical data processing in business analytics use. They would be very surprised had they known that all this time they could have 20 microseconds latency with their spark cluster.
Somehow I feel native English speakers tend to talk so much while give so little, of course not everyone, but I don’t see this happen a lot for none native English speakers
He speaks fast but yet it's crystal clear to me. Thank you.
UA-cam is kinda my hangout place. And this is one of the best videos I have watched this year!
This guy is great. by the way he talks so fast for a non-native speaker I had to rewind a couple of times.
There is a button in youtube for speed control. Have you tried that?
That was really good to hear. I have to speak at same level, so your oration here is an excellent help. Thank you.
This is the most understandable video I have watched on this topic.
Best and brightest of us...
Kevin, Thanks a lot for the talk, it was very informative.
my pleasure - glad you liked it
If you think about it, every system is (broadly understood: often the users or even developers are part of the system) taking in facts about the world, doing some sort of analysis and then taking action.
this one simple principle will build you a career in any engineering field you choose.
This is a great talk. Thanks, Kevin. At 27:23, are the Backup OR's literally on standby? That is, are they not processing requests at all? If so, what impact have you seen from transitioning to the Backup OR's? If not, do you ever run out of memory on a single machine? Furthermore is state shared across machines in this case?
As they say , to gather the crowd you must have low quality .
This quality content is high quality ,i dont know whether its good or bad for you
Great talk! Thanks
What is meant by warm up ? ELI5
Its about cache warming up aka you want your software and data to be in processor cache L1/2/3 not in the main memory like RAM which is very slow
Nice job. My 2 cents: If you're not measuring wire latency, and you are committing a lot of complexity to measure microsecond latency, than really your RTT latency is in the milliseconds, probably 5-10 milliseconds if you are cohosted at the exchange. Better focus on measuring and optimizing millisecond latency and ignore the microseconds until you have some spare time and budget.
In todays trading infrastructure, there is nothing that takes millisecond, if you are that slow you lost. Trading on FPGA are measured in tens of ns, algos right now hover around 1 microsecond.
@@dota2tournamentss The weird irony is that different market participants coexist that are working at typical latencies that vary by 10,000 to 1. The fastest three sell side equity shops are much slower than the best buy side , yet also 100 times faster than small regional banks.
Will you have anymore speech in NY
Nothing planned for the rest of this year - maybe next year? What topics are you interested in? I can add that to the list of possibilities...
@@kevgol0 hi thank you so much for your response I am a recent graduate who just wants to learn about low latency java development, I haven't found much online other than things similar to your video and quickfix j manual
@@nayemalaboni8318 Totally understand - the videos are pretty sparse. In a few days I can try and get you a list of stuff that you can watch
Hi thank you so much for your kind words and willingness to help I am truly very thankful to you for that
I am on UA-cam all the time. And, this is one of the very best videos I watched this year!
good
You nailed it !!!!!!!!
Very bad answer about security issues. But great video and nice job presenting
This is lunacy. How can you talk about HFT and distributed computing in the same sentence? The latency of communication between two neighboring computers is cosmic compared to the numbers this guy is talking about (20-25 microseconds). And in the first part of his talk this guy talks about kappa-architecture (!) and data lake storage represented by a database (!!!). That's what people from analytical data processing in business analytics use. They would be very surprised had they known that all this time they could have 20 microseconds latency with their spark cluster.
You need more substance in your talk.
Somehow I feel native English speakers tend to talk so much while give so little, of course not everyone, but I don’t see this happen a lot for none native English speakers
Java? Really?
C++ is dominating this industry.
If I had choice I'd use C only
yep yep yep
Speaking too fast causes too many mistakes, it gets quite annoying really fast.