Connectivity schematic from UltraFX VPS to LMAX Broker and Exchange Services

Electronic components do not like heat. For trading servers, efficient cooling is crucial to achieve constant low latency operation and reliability. However, for virtual and dedicated server providers the main expense is the cost of the space in the data centre. Therefore, narrow 1U rack height or blade servers often used to place many as possible servers into the rack cabinet to save space, reduce the cost and finally offer a cheap VPS. With such high density designs to achieve the superior cooling is very challenging or not achievable.

With UltraFX VPS servers we do not save space and cost to offer cheap VPSs or dedicated servers. Except the Economic VPSs we use either a 2U height rack servers or lately our own super-efficiently cooled 1.5U height Pink Jumbo® design, where massive airflow can pass through the server without any obstacle. This is the type of servers what traders and brokerage firms need to accomplish a fast and reliable operation.

Pink Jumbo® servers in our test lab before deployment in Equinix LD4 data centre:

Pink Jumbo® servers under the “funny fans test” demonstrating the high suction capacity :-) Massive amount of heat can be extracted immediately from the servers, keeping the electronic components and the CPU under a constantly low temperature:

Speed mattersOn 10th February 2016 we completed a short latency test using the LatencyTester V2 EA.

The EA measuring the time between sending one order and receiving the confirmation. The process automatically repeated 50 times to get the Minimum, Maximum and the Average latency.

The result

Min. Latency: 15 ms
Max. Latency: 32 ms
Avg. Latency: 26 ms
VPS: UltraFX VPS Premium Gamma package
Account: Live LMAX Exchange MT4

LMAX Live MT4 latency test 2016 February 10

Conclusion

UltraFX VPS clients have a latency edge compare to the majority of MT4 EA users. UltraFX VPS clients EA’s often already in and out of the trades until others still trying to catch the quotes or waiting for confirmations. We are not surprised that many of our clients reported that their EA came into life on UltraFX VPS. Even with MT4 the latency improvement clearly visible when it placed into a professional HFT environment.

blocked ipWe are not blocking or speed bumping the connection to any brokerage firms or execution venues. Our ports are available to connect to wherever you want. If you are unable to connect, please contact the brokerage firm to whitelist your UltraFX VPS or dedicated server IP address. We provide each VPS or dedicated server with an individual, dedicated IP address and therefore brokerage firms can easily whitelist only your IP address independently from all other UltraFX VPS clients.

 

However your whitelist request may will be declined by the brokerage firms and in some instances the brokerage firm will recommend a so-called “preferred” VPS provider. In such cases you should consider the followings:

• In the retail forex markets brokerage firms most of the cases act not only as agents who transmit your orders to an execution venue but in addition act alone as a venue of execution. This means that they are standing not on your side but opposite to you. So it is not the best idea to use a VPS provider recommended by a party which is not neccesarily acting in your best interest.

• Scientific studies say that: financial markets have two important functions for asset pricing: liquidity and price discovery for incorporating information in prices (Maureen O’Hara. “Presidential address: Liquidity and price discovery” Journal of Finance 58.4 (2003).
Lower latency has a positive effect on price and liquidity discovery and not least on transparency.

Most UltraFX VPS products are very fast and that why not welcomed by some brokerage firms, probably because they don’t want to share the interest of better liquidity and price discovery and transparency with you, or maybe there is not even real liquidity by them what could be discover. Therefore, if a brokerage firm does not allow you access to their trading facility just because you use UltraFX VPS virtual or a dedicated server then it is better to avoid those brokerage firm. The likelihood is you will not be happy how the brokerage firm will execute your orders or the brokerage firm will not be happy for your trading flow.

• The UltraFX VPS service provides low latency hardware and networks to benefit our clients, but on the other hand we do not provide any advice about trading strategies. In other word UltraFX VPS service is a simple hardware, hosting and network service and we do not advise or encourage clients to use any strategies which may be liked or disliked by the brokerage firms or the execution venues. Put simply, we do not involve ourselves in any battles between trading clients and their brokerage firms regarding execution and service quality. Any such issues must be settled between the brokerage firms and their clients to find a common ground, agreeable to both parties.

MT4Most people test latency with PING messages. Unfortunately PING messages give only rough measures about the latency on overall systems and most of the time misguiding about the true latency. First of all, trading messages using a completely different message protocol with different demand on the network and on the hardware infrastructure. With PING it is like sending a little ball through a big enough pipe but in reality much bigger numbers of balls need to be handled and sent through the pipe.

Overall latency determined by:

  • How fast the messages (usually TCP/IP and not ICMP like the PING messages) travel through the network between the two destinations. Here plays an important role the throughput capacity of the network which cannot be accurately tested with the few bit packets of the PING messages.
  • How fast the trading logic can complete its calculation, which is software and hardware dependent and also cannot be measured with PING.
  • What delays imposed by the other party during execution. How fast is the core execution and how fast will the confirmation about the status of the orders etc. be sent back.

For example far from accurate but it is a more sound technique to test the real MT4 setup latency with the method explained on the video below than sending PING messages. (The Latency Tester v.2 can be download from here.)

Risk warning!

Make sure that after completion there are no open orders or positions left in the MT4. In case you wish to make the test on live accounts than only do it at your own risk.

time is moneyFX market is considered slow compared to the equity exchanges. Especially from the retail FX trader perspective where real latency competition often exist only in a limited fashion. However if we analyse a no last look top of the book (TOB) prices on one of the Equinix LD4 located FX venue than we get an interesting picture.

We have measured the EURUSD TOB price life times (PLT) of 7 days (UK time from 8 am to 5 pm). The number of changes were around 12 million for this period. On the graph below, the vertical axis shows the number of new prices and the horizontal axis shows the life time of prices in milliseconds. The Price Life Time is determined by the cancellation or by the arrival of a new price onto the TOB.

PLT bid all

The table below shows how many percent from all TOB prices fall below certain PLT.

Percentage

Table explanation: 10% of the 12 million top of the book prices have a shorter life time than 2 milliseconds, 20% of the 12 million top of the book prices have a shorter life time than 3 milliseconds... and so on.

The graph and the table above not only gives you insights of the well-known presence of high frequency algorithmic market making of liquidity providers, but also provides information to the liquidity taker algo traders about the latency task they face by choosing the technical setup for their algorithmic trading system.

For example, if the algo trader runs his trading system from a Dublin located server, which has approximately a 13 ms roundtrip latency to London, to take the top of the book price on the Equinix LD4 located venue, then more than 40% of the prices will be as good as non-existent for his trading system! It will take approximately 6.5 ms until the price reaches his system and then another 6.5 ms to reach his order to the venue, but at this time more than 40% of the prices are not there anymore. It is absolutely clear the importance of colocation. However colocation is not the only important factor. Fast VPS or dedicated servers with a low latency network cross connected to the execution venue further increases the efficiency. For example, if the trader’s round trip latency is slower than 3 ms, then 20% of the TOB prices are out of reach. Every millisecond or microsecond counts. Cheap VPS service where speed is only a marginal factor could cause so much delays for a trading strategy that at end of the day it could cost the trader much more than a higher priced service where latency reduction is the first priority.

The latency not only for higher frequency trading algos are important, but often for less frequently trading systems as well as soon orders need to be placed or modified in reaction to position or price changes. Let’s assume that a strategy which uses an hourly chart (hourly sample) gets into a position and a trailing stop gets activated on the trader’s server. In this case, it could be crucial how fast the trailing stop gets modified and placed on the execution venue’s server. Especially when the trailing distance is set relatively close to the current prices.

Latency reduction is our first priority and clients from UltraFX VPS are often in and out of a position in a time when other trades only begin to react.

LondonThe foreign exchange market is actually a two level market. The top level is the interbank market and the second level is the non‐interbank FX market. Major banks and market maker firms provide quotes for the non‐interbank FX markets. They have an order book, which will go net long or short during the execution of the non‐interbank FX orders. Every liquidity provider has a predetermined level for the maximum size of this net long or short position in each currency pair. When this level is reached, the liquidity provider will reduce the long or short exposure by buying or selling on the interbank market. This buying or selling still mainly occurs on the ICAP EBS or on the Reuters dealing system. If liquidity providers filling and taking retail orders into their books, which are worse than the actual prices of ICAP EBS and Reuters dealing systems than they will suffer unwanted losses. Therefore prices at these two venues act as a reference price for the liquidity providers. ICAP EBS has matching engines in different cities but only London is the only place where the matching engines for both ICAP EBS and Reuters dealing systems can be found. Because of this, London is a unique place regarding forex execution. To receive prices from these venues with the least delay possible, is a crucial element in liquidity provision for the non‐interbank FX market. Here, the importance of colocation comes into play because the closer the liquidity providers are to these two venues, the faster they will receive the reference prices. Non‐interbank forex traders also need to be closer to these venues because most likely, they will see better executable prices coming from the liquidity providers. The Equinix LD4 data center is the best place in this regard and most liquidity providers send quotes to the non‐interbank European venues from there. It is often said ‘everybody who counts in FX business has a presence here”... It’s no accident therefore that UltraFX VPS and UltraFX Dedicated Servers are located in London (UK), in the Equinix LD4 data center.

Safety firstBesides all the benefits that computerized automated trading offers, there is additional risk involved.

In any situation where machines take part in a process, there is a possibility that the machine malfunctions or even fails completely. Traders have to be aware of this risk and have to be prepared for unwanted events as much as possible.

The first and maybe the most important rule is do not leave the automated trading system unattended. We have heard stories about people going to work or on holiday and at the same time their robots trade alone. This type of algo trading is very irresponsible and dangerous. Human is the last result of safety, who can press the “big red stop button” to end the process. Therefore there must be a person there who can monitor the systems and intervene when necessary.

Automated trading systems are getting faster and faster and today even the lower tier latency systems are fast enough to cause substantial damage for its operator, in a short period of time. Users of collocated trading systems (like our VPS users) need to be especially prepared, because the physical distance makes it impossible to “quickly pull the plug”. It’s important to know that the low latency trading machines main objective is to be fast and the second objective is to be highly durable. It is impossible to build a racing car which is at the same time capable to be used as a truck to transport heavy concrete blocks. For example our servers and our network environments have a certain level of redundancy, but we designed it mainly for speed and not for safekeeping the stored data. Such highly tuned systems are more likely to have interruptions from time to time. However we are happy to say that our systems have run on a 99.99% uptime from the beginning.

  1. The base camp of automated traders is their servers or PCs running usually at their office or at their home. The collocated VPS is the frontier “fighter” which has more chance to fail in the battle. Therefore all strategies, important information, etc needs to be kept safe on the base camp computers and not on the collocated VPS. The only things that should be installed on the VPS are those essential to perform the trading operation.
  2. From the base camp computers traders must have a VPS independent connection to the broker (or trading venue) to close positions and if necessary completely block the trading access for the algorithm installed on the VPS. If blocking the access for the algorithm running on the VPS is not possible through the base camp platform, then direct telephone contact with the broker needs to be an open option.
  3. We highly recommended that the base camp has a minimum of two internet connections from different internet service providers. It has to be set on a fail over mode. When one line fails, then the connectivity is taken over by the other one. Many individual traders use the landline internet connection as their primary and for failover their telephone 3G internet line which is usually good enough to perform emergency tasks.

Emergency procedure.

When something goes wrong and hard earned money is at stake emotions often take over. Decision making is then rarely logical. For such instances every algo trader must have their own pre prepared emergency procedure. The procedure is a check list of “what to do next” a step by step guide written on paper (not on the computer in a word doc). It could be a flow chart. This paper should lie on every algo trader’s desk but even better, stuck on the wall.

Example for the event when the co-located VPS or dedicated server is not accessible:

No VPS access flow charts

Preparedness

The above flow chart is a rough example. Everybody has to plan their own emergency procedure for the different emergency events.

It is necessary to analyze different scenarios and work out the “what to do when” answers. It is very important to have all contacts (telephone and email) written down on this paper. Practice makes the master… Therefore we recommend from time to time, to go over the emergency procedure to make sure that when you need to act, it will happen fluently.

How to test you present preparedness? Check with a stopwatch how long it would take you to make an immediate telephone call to your broker and tell him your account number. Remember every second could cost a lot of hard earned money.

direct link

No ballooningThe first and obvious advantage is that our servers (VPS and dedicated) are in London (UK), in the Equinix LD4 which is the best place that a forex trader can have. Furthermore, we provide servers purely for trading applications and our systems are specialized to fulfill the requirements posed by trading algorithms. Hosting websites and storing family pictures need different requirements and often contradict with those needed for trading. We run our UltraFX VPS business as a Club of Algorithmic Traders and we know that our clients as members of the club demand high quality systems. So we have to constantly improve, especially the latency numbers, just to keep this club alive. We use high grade servers. From the middle of April, 2015 we switched on our ultralow latency internal network and new faster products are now offered as well. We never do „ballooning” which means we distribute only the amount of RAM capacity that we physically possess. This ensures stability and reduces the chance of disruption of service caused by resource exhaustion. Approximately, we host only half of the number of VPS that we actually could on one server, to make sure that CPU is available when we need it. We know how secretive traders are about their trading systems and because of this we give full administrator rights with all VPS and dedicated servers. When the trader changes the password given by us then even we are unable to access his VPS or server. We can stop it, restart it and erase it but we’re unable to access to the stored information. Speed and privacy are paramount overall.