BackNSE denies employees connived with trading member for faster access
NSE responses to Mint's queries on Bombay High Court order on transparency
Premium
In a recent order, Bombay High Court judge G.S. Patel lambasted the National Stock Exchange of India Ltd (NSE) for lack of transparency, among other things. Sucheta Dalal, managing editor at Moneylife, had approached the NSE for its response to a whistleblower’s complaint. Not only did the exchange not respond to Dalal, but it subsequently filed a defamation case against her and the publication for publishing the allegations made by the whistleblower. Patel said in the order that the exchange may have had a case if it had responded to Dalal, and provided that the allegations had been published without further investigations, as if no response had been received.
Mint received the following emailed responses from the NSE on queries regarding the order:
a) With reference to this statement made by NSE to the high court—“The time lag between the first person and the last person (depending on the load) is maximum 50 micro seconds (1 micro second = 1 millionth of a second) which by no stretch of imagination can confer any advantage to any person"—can you confirm if this time lag was occurring under the previous transmission control protocol/internet protocol (TCP/IP) model of communication? Is it true that under the earlier TCP/IP model of communication, the first trading member to connect to NSE’s servers was the first to receive data feeds, even if the time advantage was in micro seconds?
Yes, this is true for the maximum difference between first and last on a particular server port when the data leaves the server. However, the first member will not necessarily receive the data first, since there is network congestion to deal with, which is usually 5x to 10x slower. The best example is that two people may leave a station in a car; Person A leaves a few seconds before the Person B but there is no way to know who will arrive at the destination first. If on a train, or if there is no traffic, then the person to leave first arrives first. Not so by road, which has a large variance in travel times. To ensure Person A arrives first, he might have to leave at least some time before Person B (even more during peak hours), not a few seconds. A network is like a road, with congestion, multiple routes and variances (called jitter).
b) Was there (at any point in the past) any manual intervention involved in the data connectivity as alleged in the whistleblower’s letter to Securities and Exchange Board of India (Sebi)?
No. In fact, members are allocated ports and server IPs during registration and continue to use the same IP and port from thereon. They are not reallocated either manually or automatically. Logs clearly show that members have not moved IP or port, except on rare occasions (once or twice a year) when new servers were added or problems happened with an existing server. The connectivity and server reboots are handled by an automated script. The process as a whole is started manually, but there is no scope for intervention in the individual steps.
c) When did NSE migrate to the UDP (User Datagram Protocol) model of communication? Was the migration because of concerns expressed by co-located trading members about differences in latency?
NSE uses IP Multicast, not UDP. We offer IP Multicast as an option—many of our members continue to use TCP/IP. Technologically, the primary reason is efficiency. IP multicast uses significantly less bandwidth, but is more complex to monitor and more prone to failure, which thus requires more error-correction and recovery. Many members prefer simplicity and higher reliability to bandwidth efficiency, so continue to use TCP/IP.
d) We are given to understand that within trading members who are co-located, the servers are located in three separate locations at NSE, as co-lo racks were made available in three phases. There are concerns among trading members that this, too, leads to differences in latency. Is this true? If not, what are the measures taken by NSE to ensure latency for all co-located members is the same?
The only difference in the phases are the options available (1G/10G, full or half rack, etc). Latency is identical.
e) Has the exchange taken any disciplinary action against any employee—such as termination of services or transfer to another department, etc.—owing to connivance with any trading member(s) for faster data access?
There was no such instances, so the question does not arise.
f) Sebi’s circular in May this year said that exchanges should provide fair and equal access to co-located trading members. What are the metrics shared by NSE with the regulator to show that access is indeed fair and equal? Has the nature of these metrics changed after the recent circular?
Sebi has asked exchanges to provide fair and equal access to colocation space, so that no member is prioritised, refused access or charged differently. We have ensured that. NSE has always ensured first come first served and that there is enough colocation space for any member who wishes to access it. NSE continues to plan capacity in advance of demand.
g) Following up on the previous question, the Sebi circular in May said:
3. In order to ensure fair and equitable access to the co-location facility, stock exchanges shall...
...3.2. ensure that all participants who avail co-location / proximity hosting facility have fair and equal access to facilities and data feeds provided by the stock exchange.
3.3. ensure that all stock brokers and data vendors using co-location / proximity hosting experience similar latency with respect to exchange provided infrastructure.
This goes beyond mere access to space in my reading. And gets into the sphere of experienced latency, etc. Your response?
The circular talks about access and latency. Latency is ensured by giving all members the same route and all colo facilities the same wire distance from the server. Latency is not priority – it is time between when the message leaves and when the message arrives. As mentioned earlier, this actually has to do with the network architecture (distance, number of network nodes), and not to do with which message leaves first.
A brief explanation to some terms used in the Q&A above:
1) The TCP/IP (Transmission Control Protocol/Internet Protocol) model of communication scores high on reliability, but also results in minor differences in the speed at which market data reaches each trading member. The trading member who is the first to connect to the exchange’s servers is the first to receive market data feeds till the time the connection is alive. According to the exchange, this may not be necessarily so because of so-called jitter; although technology experts say jitter should hardly matter in data travelling in a short distance to co-located servers.
Among other allegations, the whistleblower said that a trading member connived with an exchange employee to gain daily information on when the exchange’s servers went live, and then become the first to connect to the server. The exchange has denied anything of the kind occurring.
2) The exchange introduced IP Multicast, where data is sent simultaneously to many recipients together, a couple of years ago. This is seen by some market participants as a response to criticisms about time differences in receiving data feeds under the TCP/IP model.
While the multicast model addresses the issue of time differences in data feeds, it scores lower than TCP/IP on reliability. However, trading members who are willing to make large investments on technology have migrated to this model of receiving data. As the exchange’s response above says, using this model “requires more error-correction and recovery." Members who prefer reliability continue to use TCP/IP.
Unlock a world of Benefits! From insightful newsletters to real-time stock tracking, breaking news and a personalized newsfeed – it's all here, just a click away! Login Now!
