Which broker offers lower latency—Zerodha or Dhan—when it comes to receiving real-time market data and executing orders? I’m working on a strategy where even a 50ms delay could be critical. Additionally, which of the two provides more reliable market data?
I came across a claim from Dhan stating that 90% of its orders are executed within 25ms. However, I couldn’t find any similar benchmarks or official data for Zerodha. Does anyone have insights or real-world experience on this? And how accurate is Dhan’s claim?
I’m aware that NSE colocation is an option for ultra-low latency, but it’s currently too expensive for me
While we wait for folks having experience with various brokers’ APIs, to share their observations/anecdata, here are a few adjacent thoughts on the other aspect…
…and how relevant/actionable the claim might be
for a trader looking for low-latency, high-frequency trading support.
IMHO, you will be better off evaluating the end-to-end latencies that your trading setup/infra faces when connecting with different brokers and executing your specific strategy, The numbers posted by any broker will unlikely be your effective latency, on your infra, for your strategy.
This is especially critical when one is NOT hosting one’s algo on the broker’s infrastructure, i.e the broker will absolve themselves of any latency involved between one’s trading setup/infra and the broker’s infrastructure.
Reminder that this does NOT mean that,
of every 10 orders from each user, 9 will be processed within 25ms.
Without knowing the distribution of the latencies involved, and the period over which such a claim is made, i doubt such claims are worth much to build a low-latency trading strategy upon. Especially when such a metric is not even being guaranteed.
Apparently, NSE routinely handles millions of daily orders (even billions?).
Consider that, any broker claiming aggregate vanity metrics like “90% confidence of 25ms latency, and unspecified latency otherwise”, can achieve such a metric even after delaying way beyond 25ms ALL their users’ thousands of orders during a critical market frenzy period, by handling 9-times as many orders during the rest of the day within 25ms.
FWIW, earlier this year Dhan had claimed “95% orders (in a month?) within 25ms”.
Also, when arriving at such metrics, there are various assumptions made. The discussion in the above link contains a few such examples and their actual impact on certain strategies/trades (additional latency, race-conditions between orders) .
Basically, depending on the strategy one wishes to deploy, there may be other bottlenecks in a broker’s tech-stack that can result in a higher overall latency, or differences in the broker’s implementation that can result in a difference in execution.
…so best to actually deploy one’s strategy (on a small-scale?) and evaluate how various brokers perform, and adjust the strategy or the choice of broker appropriately.
Alternately, one might need to ensure one’s strategy has alternate risk-management in place to account for and handle the scenarios when such an expected latency is NOT achieved by a specific order (or a set of orders), to ensure that losses (if any) are limited.
Also, location of your server and their servers would impact latency. If you are in the same datacenter vs if you are operating your API client from somewhere far away.
While Zerodha’s Kite Connect API has been instrumental in opening up algo-trading to a broader retail audience, it’s important to acknowledge that its current architecture is not well-suited for latency-critical or high-frequency strategies. Orders are routed over internet-based leased lines and processed in a FIFO manner, which—combined with Zerodha’s large user base—introduces inevitable latency during peak hours. There are no published benchmarks or guarantees around order execution speed or data latency, making it difficult to design or trust latency-sensitive strategies. Moreover, the absence of co-location options or dedicated low-latency infrastructure puts retail traders at a disadvantage compared to brokers who offer these features. While Kite is excellent for discretionary and moderate-frequency algo strategies, traders relying on sub-50ms execution need to realistically assess these limitations. A more transparent stance from Zerodha on latency metrics and potential future enhancements would be highly appreciated by the growing community of serious retail quants.
I recently tested the order placement speed of Upstox’s API. Here’s how I conducted the test:
First, I identified a stock where only one quantity was available at a specific price in the order book. Then, I simultaneously placed a buy order for that single quantity through both platforms’ APIs. Whichever order executed faster would fill the order, while the slower one would remain unfilled, since only one quantity was available.
I repeated this test multiple times. The results showed that Upstox’s API was faster approximately 55% of the time during regular trading hours and around 90% of the time during high-activity periods, such as market openings, closings, or during significant price movements.
I can switch to upstox, but I don’t know if it is stable. I am using Zerodha from few years, and it seems stable 99.5% time. So I won’t switch to upstox till I try it for some months.
Note: I tested with Upstox plus plan, where brokerage is 30 INR instead of 20 INR. Since it is faster according to them.
We have an average response time of ~20ms internally between our OMS and the exchange. However, since clients connect to us over the internet, there is no way to claim any response times. It is completely dependent on the internet line.
Any strategy that is sensitive to 50ms latency should not be over the internet; it should be colocated because over-the-internet latency cannot be controlled for.
We have dedicated leased lines between our OMS and the exchange. Over-the-Internet connection is between us and the client, which is true for all brokers.
Adding to Matti, we have multiple instances of setups means we distribute clients among many silos so each silo is like a separate broker setup, we maintain multiple of them so the concept of huge user base won’t arise.
If any retail broker is claiming they have best latency we need to take it with a pinch of salt, because the whole retail architecture is such a way where multiple vendors and multiple node points are present thus each one introducing it’s own latency, for example leased line, internet, cloudfare etc.
@siva@Matti
Could Zerodha introduce a “premium silo,” executing orders exclusively from a limited number of clients for a charge of XYZ amount? This would allow clients to benefit from a faster API.
Although my strategy isn’t ultra time-sensitive, a quicker API directly translates to higher profitability. Thus, it would be a win-win scenario for both Zerodha and its clients.
@siva does zerodha have anything like “premium silo” just like you have some separate basket system for traders who place arbitrage or spread orders on a regular basis.
Agree with Siva. We can offer a dedicated silo for some users and charge them more, but we still can’t guarantee over-the-internet response times even there. So it will not deliver any real value.