There are a number of articles on the subject of grid and cloud computing. There’s also a lot of attention on companies like Amazon Web Services, Google, IBM, Sun and others who provide services in this space. For the purpose of this article grid computing is defined as the process of distributing a single or multiple tasks across many computers at the same time. The premise being that something can be done N times faster running on N computers simultaneously. Let me take you though my thoughts on how a hedge fund can benefit from grid computing.
This you may think is for the domain of academia, research, big banks and others with the resources to build and manage these super computer networks. At present there are many applications of grid computing and it’s nothing new. It’s been around for as long as someone could connect one computer to another and do something in parallel. Your everyday life touches on things that at some point during their development has benefited from grid computing; weather forecasting, medical research, special effects in movies, the tread pattern on your car’s tyres and many more.
To bring grid computing closer to home let’s look at an example of a hypothetical hedge fund that employs around 30 people. From the start, they outsourced as much as possible to their Prime Broker and Administrator. After their first year of trading, they decided to take some of their processes in house for greater operational control and to implement a portfolio management and risk system for which they purchased three additional servers to accommodate their new systems’ processing and reporting.
With greater control it was now possible to extract and process a myriad of new reports and analytics resulting in the speedier processing daily processes and management felt better informed for operational and investment decisions. However in the background an invisible cost was starting to grow. As part of the systems implementation they employed an extra person to take care of the day to day overhead of running the in house systems. This person was assisted by existing IT support staff as well as some ad hoc support from the system vendor. Due to an ever growing transaction volume, transaction history and more reports to produce, the end of day process was taking 3 hours to run compared to the 30 minutes when the system was first implemented. This did not matter that much as they had the whole night to schedule and run these processes.
Throughout this time the businesses demands on the IT system grew; for example the company email database was growing and more and more documents stored on the central fileserver. There were now clear signs of growing pains. Disk drives regularly filled up, the database server’s performance slowed down. This all resulting in longer processing of end of day and slower responsiveness of the trading system’s front end applications. Much of this overhead was initially swallowed by the company and written off as the cost of having one’s own systems.
This progressed to failures multiple times a week. The end of day process would be re-run during the day on T+1 and PL and Risk reports were only reconciled and available at the end of the day on T+1. In order to address these problems, they decided they needed to upgrade their IT infrastructure, purchase additional servers for the operational processes and hire additional support staff. The management signed this off as a result of a growing business.
Does this situation sound familiar to you? I would guess that either you or someone you know is going through this right now. So what has this to do with grid computing? The picture painted above is not that hypothetical and has been constructed from various personal experiences. The essence of this problem is scalability. Every one faces this issue at some point; how do you implement a solution to meet immediate requirements which will also provide a solid enough platform from which will accommodate growth and future needs.
There are a host of books and online resources one could research for assistance in planning and equipping yourself for growth. Till recently smaller business willing to invest scalable infrastructure had a hard decision to make. Make a substantial capital and operational investment into technology and processes that would initially be substantially underutilised and in the future run the risk of not been properly suited. Due to the very high initial investment and uncertainty of return this has been a very unpopular approach.
Additionally, the team responsible for managing technology in a typical small to medium sized hedge fund doesn’t have very much to go on in terms of estimating future growth requirements. In most cases they don’t have the budget to deploy infrastructure and systems that would cater for unknown growth, in particular very fast growth.
The good news is that recently there is more choice for those wanting to be better prepared for these uncertain times we live in. The evolution of the Internet has pushed the development of various scalable technology solutions. Particularly since the bust of the Dot Com era companies have had to assemble technology that is economical, functionally rich, highly available and scalable. One of the beneficiaries of this development is grid computing and I will show how this can addresses the processing requirements for a hedge fund’s operational systems and processes.
When you purchase additional servers to solve processing problems it always comes with a cost. Where do you store them, electricity, cooling and how best to administer them? Next you have to decide how most effectively use them and to share processing across them.
In most alternative management businesses, the bulk of the computer processing is over night and for the remaining time the computers are relatively idle. How can one balance this? How do you spread your computer processing evenly over 24 hours to get the highest return from your technology investment? Now back to our example hedge fund with 30 employees. They would have at least 30 workstation computers of which mostly are relatively idle. For the majority of users in this organisation they have a workstation that could not ever be very busy or at least only fraction of the time. Here resides a computing resource of amazing potential. Workstation computers today are configured with hardware components not far off of what would be used in servers. In actual fact they are mini servers. Now imaging each of these computers can communicate with a central server, request a work item that would run in the background with the user unaware and unaffected. With this resource why would our example Hedge Fund want to purchase additional servers?
Implementing a grid can turn out to be fairly complicated when you get into the detail of how to co-ordinate and run processes across this available computing resource. There are a number of open source initiatives and commercial products that set out to solve this problem and make grid computing more assessable.
Let’s look how a hedge fund can implement a grid computing solution using the ScaleFast products Grid and Flow.
ScaleFast Grid comprises of a central application that manages individual work items to be executed. A number of grid agents or workers are installed on all computers that at some time will be available to the grid. These workers understand how much resource they are allowed to make use of in the host computer. ScaleFast Flow is a process scheduler, it know how to bring together and co-ordinate business processes that are made of up dispirit programs, scripts and user interaction across the business.
Now we have a tool which can help us, what next. Firstly the business must define its process requirements and service level agreements (SLA’s) for each process e.g. Provisional PL report for fund must be completed and distributed to fund managers by close of business on Trade Date (T). Risk Reports based on overnight positions, must be ready first thing every morning. Profit and Loss , position and trade reconciliation completed by midday T+1. NAV light, completed 15:00 T+1.
For our example we have set certain ground rules from which everything stems. Next we need to look at where our problem areas e.g. what take a long time to run, where do things often fail and what bottlenecks occur because everything currently runs one after the other.
ScaleFast turns you available computers into a single platform made up of multiple computer slices where many tasks can be run at the same time. It could be possible to run the risk report and the PL report at the same time, further it could be possible to split these reports up into smaller pieces. Assuming the Risk and PL reports take the same time to run, by only having a grid of 4 slices or workers would result in processing time improvements by over 400%.
ScaleFast Flow adds another dimension of performance to this example. It versions and controls the execution of processes for a number of tasks that have various interdependencies. If something goes wrong in this process you may not want it to proceed, after the problem has been resolved you may want to resume it from a certain point. You may also not want to have to pull in an IT resource every time something goes wrong, and you may also require a full audit trail of when and why a process failed.
Take for example this fairly common process. Snap market prices -> Fetch market data from broker -> calibrate curves and volatilities -> send trades to administrator -> Run Risk Reports, Run PL reports, Run Recon extracts.
Some steps may not be critical to the process, some steps may be executed in parallel, some steps have to run on certain preconfigured servers and some steps would benefit from been automatically split into smaller more optimum steps. With the ScaleFast tools this is easily achieved , users have experienced process optimisations from eight hours down to 40 minutes.
Grid computing gives the hedge fund a scalable platform on which to manage and grow their internal processes. With the high market volatility so common these days more intra-day real-time reports are becoming an essential business tools in managing the trading and operational risk within a hedge fund. ScaleFast provides a functionally rich and cost effective platform for those requiring more from their systems and faster results.
Robert Betts is the lead consultant and CEO of Develops Ltd, an Alternative Investment Management consultancy. Robert speaks and offers commentary on various technology related topics relevant to the industry. Develops has a strategic partnership with ScaleFast, a software and services company making grid computing accessible for main stream use.
No comments:
Post a Comment