Whats best for serous ATS: lightspeed vs rithmic vs DAS

I design infrastructure for a living for front, mid and back office systems... removing latency, is part of it... and the ULLT space is only reserved for those strategies that make markets or have an arb edge... so please understand... $100K will still not get you ULLT to compete against a bank or fund...

your current strategies.... how would they benefit from being closer to each exchange venue? that is what you need to ask yourself first... then second, if you are coding your own platform... whose OMS are you going to use? because all that you mentioned are prop or retail and those are a joke for the most part if you are trying to go ULLT...

I think you are approaching this problem the wrong way, perhaps because you are an app architect with no experience on this field; but you should know that infrastructure is based on the app req's and as such your focus should be first on what your strategy req's are for which you will design the app and go from there...

There are hw vendors that tune things for ULLT, so dont be too worried about that aspect. Linux is the de-facto standard for it, while I know funds who are also doing LLT(not ULLT) on Windows in the cash equity markets... so again, what is your strategy, what does it require to generate alpha...

there are books that cover this thought process... go to amazon or safari or google books and give a few a read...before you invest tons of $$$$...

Quote from ptumati:

ofthomas & Swan Noir

Yes, as you have suspected, I do not have much trading experience and yes, I am a newbie. Here’s what I am trying to accomplish:

Understand ULLT/colocation and use this knowledge to design a multi-strategy (strategy agnostic) trading system first. I shall start with 80-90K and run medium frequency trades(50-100/day) with a couple of strategies I have at hand. If PNL looks good after a few months, I will then collocate the system go for higher frequency strategies on the same trading system. All the discussions centered around ULLT/software… etc is to make sure that I take informed decisions and design the system in such a way that I don’t have to redesign it again whenever I get to the colocation stage… in my view, a multi-strategy system that works for medium frequency trades should also be able to process faster frequency provided we change the strategies and the data feeds. I hope I am thinking clearly on this.


Yes, this is the problem when one enters a new field… its hard to figure out how much we know vs how much we don’t know and so these questions :(
More...
 
Quote from jb514:

IMO, don't put colocation on such a high pedestal. It's not as expensive as people think, just do it.
More...

I can colocate for $100 a month per U and be 0-1ms latency from my market data and order entry servers... like I am today... but colocation is only one portion of the cost... the point is that it is not needed for every strategy, I colo only because i care about uptime and reliability of my connection.. something I cant get in the burbs of CT...
 
Quote from ofthomas:

I can colocate for $100 a month per U and be 0-1ms latency from my market data and order entry servers... like I am today... but colocation is only one portion of the cost... the point is that it is not needed for every strategy, I colo only because i care about uptime and reliability of my connection.. something I cant get in the burbs of CT...
More...

So this is the more casual setup ... I'll bet your "order entry server" is a middleman who routes it off to the exchange. Have another middleman who consolidates and passes you the market data. The data feed is the likely bottleneck (know where your data feed is connecting to ... does nxCore have a server in NY yet, or still connecting to Nebraska?) With this setup you pay $300 - $1500 month for technology and data, but more like 5-10ms at best (from the time the quote changes to the time your order hits the market)

The hardcore setup is subscribe directly to exchange data, consolidate your own data (and this is where you want to look at some serious hardware and optimizing your Linux kernel), and send your orders directly to the exchange ... and of course the exchanges know you are a serious player and it costs tens of thousands of dollars a month.

So Rithmic has a great connection to BATS .... if you are adding liquidity, that is great, you can post your order in 250us. And in my experience, posting orders to BATS is as good as or better than posting to whatever native exchange (NASDAQ or NYSE). Actually my fills are statistically better posting to BATS (but, mostly with 100 and 200 share orders, so take it with a grain of salt). The liquidity will find you.

But the caveat .... if you are taking liquidity, chances are the NBBO isn't on BATS and they will re-route your order anyway, so now you are talking many milliseconds, not microseconds.

Anyone can give you "best case" numbers, but you need to look at the round trip time, from the time of the quote to the time your response order hits the market. The data source is likely to be the bottleneck. Know how many hops you are from the data source (exchange) and what kind of hardware they have for doing quote consolidation. Lime and Rithmic are integrated solutions, so you get fewer headaches and great speed at a reasonable price ... I don't know much about lightspeed.
 
Ok, now, I see three different setups here:
1. Casual setup with $300-1500/month (inhouse/colo+datafeed etc) tech cost that at max can get to a latency of about 5-10ms from Trade TO Order placement ( say TTOO delay) as promagma mentions
2. Hardcore (LLT) setup below 5ms hedge funds etc
3. Ultra Hardcore (ULLT)…. used by the gods & kings

Based on the feedback, options #2 & #3 are off. So, #1 is what I should be looking at. I feel that I can reduce the TTOO to less than 100ms:
Time from trade to nxcore 5ms (assumption)
Travel Time from nxcore to my place (ping www.nanex.net/2) 8ms (*) [Comcast 19mbps download]
Time to process data and prepare order 1ms
Time from my comp to rithmic/lightspeed 30-8ms (*) (30ms rithmic/8 lightspeed)
(Ping rithmic.com/2 & ping lightspeed.com/2)
Time to place order at ECN 1-5ms
--------------
Total time 50-30 ms (approx.)

If, I don’t colo, I end up paying nanex anwhere between $500-$750 (Equity ONLY) and if I do, I shell out an additional $1000 (few feet from Nasdaq as lightspeed promised). Colo can save me upto 30-8ms (+4 nxcore westcoast vs datacenter); approx 5-10ms TTOO with a monthly budget of about $1500-$1750 is possible.
As for trading strategy, I have been working on two strategies that seem to be promising. They are generating descent profits (paper trading mostly), but sudden price movement is ruining everything.
As for the order routing, light speed’s datacenter is located a few feet from nasdaq and they offer a wide range of routs (See pg 62: https://download.lightspeed.com/GatewaySpecs/BDKAPISpec.pdf). So, coming to your quote:
Actually my fills are statistically better posting to BATS (but, mostly with 100 and 200 share orders, so take it with a grain of salt). The liquidity will find you.

But the caveat .... if you are taking liquidity, chances are the NBBO isn't on BATS and they will re-route your order anyway, so now you are talking many milliseconds, not microseconds.
More...
Are you saying that a liquidity adding order gets filled in quickly (because liquidity will find the order) vs a liquidity removing order that needs to be routed to a non-BATs NBBO? So, if that’s the case, Rithmic (BATS) is good for all the liquidity adding orders and good routing options provided by lightspeed are even better since they can likely route the order directly to the ecn that offers the NBBO….

"If that’s the case, lightspeed’s multi routes feature might be a better option compared to Rithmic’s one route to BATS. Of course, it depends on the strategies we use as well…"

Any comments on the above statements/assumptions/conclusions?
 
I agree with all you are saying... true hardcore talks to the MDS Servers of the exchange in question directly for their quotes, and also sits next to the sequencer servers and rebuilds the orderbook...and it costs way more than what most think, not because of a single component, but because of the aggregation of overall costs..

In any event, I fail to see how anyone on this forum, unless you are an actual fund, would benefit from doing ULLT/UHFT unless they had the backing and expertise to do so... specially because you would have to be at each exchange location or each market in which you transacted...

so continuing to discuss this subject is kind of silly given the OP lacks the access to do it the true way, at best he can compete with retail, not with funds... so in this case he is going to lower his latency to below 2ms, at best... but without even knowing how that would benefit his strategy, not sure why he is even looking at it.

Quote from promagma:

So this is the more casual setup ... I'll bet your "order entry server" is a middleman who routes it off to the exchange. Have another middleman who consolidates and passes you the market data. The data feed is the likely bottleneck (know where your data feed is connecting to ... does nxCore have a server in NY yet, or still connecting to Nebraska?) With this setup you pay $300 - $1500 month for technology and data, but more like 5-10ms at best (from the time the quote changes to the time your order hits the market)

The hardcore setup is subscribe directly to exchange data, consolidate your own data (and this is where you want to look at some serious hardware and optimizing your Linux kernel), and send your orders directly to the exchange ... and of course the exchanges know you are a serious player and it costs tens of thousands of dollars a month.

So Rithmic has a great connection to BATS .... if you are adding liquidity, that is great, you can post your order in 250us. And in my experience, posting orders to BATS is as good as or better than posting to whatever native exchange (NASDAQ or NYSE). Actually my fills are statistically better posting to BATS (but, mostly with 100 and 200 share orders, so take it with a grain of salt). The liquidity will find you.

But the caveat .... if you are taking liquidity, chances are the NBBO isn't on BATS and they will re-route your order anyway, so now you are talking many milliseconds, not microseconds.

Anyone can give you "best case" numbers, but you need to look at the round trip time, from the time of the quote to the time your response order hits the market. The data source is likely to be the bottleneck. Know how many hops you are from the data source (exchange) and what kind of hardware they have for doing quote consolidation. Lime and Rithmic are integrated solutions, so you get fewer headaches and great speed at a reasonable price ... I don't know much about lightspeed.
More...
 
Quote from ptumati:

Are you saying that a liquidity adding order gets filled in quickly (because liquidity will find the order) vs a liquidity removing order that needs to be routed to a non-BATs NBBO? So, if that’s the case, Rithmic (BATS) is good for all the liquidity adding orders and good routing options provided by lightspeed are even better since they can likely route the order directly to the ecn that offers the NBBO….

"If that’s the case, lightspeed’s multi routes feature might be a better option compared to Rithmic’s one route to BATS. Of course, it depends on the strategies we use as well…"

Any comments on the above statements/assumptions/conclusions?
More...

I am saying in my experience with small orders, adding liquidity on BATS is just as good or better than on NASDAQ. Meaning, in cases where the price just touches but does not cross my limit price, I have the same or better chance of getting that "lucky fill". For your other question, I can only guess how lightspeed's multiroute might compare to routing to BATS and then BATS reroutes to NBBO.... anyway it is probably splitting hairs, because most likely your bottleneck will be your data feed.

I use ststrader (I have been with them since they were "Genesis"). I used to be colocated and it was average 5-8ms "TTOO" when I paid for the expensive fast data feed (> $1000 mo). But now the markets are slow, my old strategy barely works, so I switched to a $40/month VPS in NY (commercialnetworkservices.com) and total cost is $350/month, but 15-30ms "TTOO" .... I am happy with it, because right now, more than speed, I want reliability to collect tick data.
 
Hmm... I was only considering coloation, but VPS sounds to be a better alternative as well. I am trying to understand all the possible approaches I can be realistic about.

But now the markets are slow, my old strategy barely works, so I switched to a $40/month VPS in NY (commercialnetworkservices.com) and total cost is $350/month
More...

If I may ask, what data feed are you using? Looking at your monthly costs, I assume that you are using one of the Active Ticks/iqfeeds of the world and if so how does the latency look like? Is the latency you observe directly proportional to the number of symbols that are subscribe to? I have seen a posting elsewhere stating that AT latency can get as high as 500ms on an average, I was not sure if that includes coast to coast transmission time. I was wondering how these datafeeds perform in one of the east coast data centers in NY? thanks.
 
Quote from ptumati:


As for trading strategy, I have been working on two strategies that seem to be promising. They are generating descent profits (paper trading mostly), but sudden price movement is ruining everything.
More...

I dont think you understand yet, perhaps is because you are a software guy... who knows...

until you fix your strategy, specially if "sudden price movement" ruins everthing... DO NOT BE CONCERNED WITH LATENCY... you are focusing on the wrong thing... get your strategy working consistently and then worry about how proximity can aid your strategy...

lastly, dont trust sim/paper trading... fills are never "true" so keep that in mind... allow for variance on the fills.. specially in equities... just because you see .01 doesnt mean that you will get fill when someone else (ULLT/HFT) beat you @ the sub-penny level...
 
You must log in or register to reply here.
Back
Top