January 28th, 2008

 

NEW YORK-Global investment bank Lehman Brothers is preparing to share its various production computing grids among several of its businesses with the eventual goal of merging them, firm officials tell DWT.

"The main technical accomplishment for this year is to become completely comfortable sharing servers and grids between applications," explains Thanos Mitsolides, senior vice president, fixed-income derivatives technology and analytics with Lehman Brothers. "Once we get comfortable with sharing them, we will merge with other business grids," he adds.

Before the various businesses can share grid recourses, the IT organization needs to troubleshoot several issues, such as continuous monitoring of the grid and making sure that all high-priority jobs get done first, says Mitsolides.

Currently, Mitsolides' group runs two grid implementations. The larger production grid, which Lehman launched in 2003, consists of 700 hosts and 3,000 CPUs running version 3.1 of Platform Computing's Symphony grid management platform on a mix of dual- and quad-CPU blade servers powered by Linux. The smaller grid, implemented in 2007 and mainly used for testing, consists of 25 hosts with 100 cores and shares the same software and similar hardware configuration as its larger counterpart. Each grid is linked logically, but their various servers are spread across three datacenters in New Jersey and one in New York City.

Within a year, Mitsolides expects to see the total size of both grids double, with the smaller grid eventually representing 70 percent of total grid resources.

To handle the file system issues of its growing grid, Lehman officials are investigating a number of solutions. "We can speed up the network file system (NFS) by using faster NetApps or by moving to a virtual distributed file system, but the virtual file system wouldn't happen for another two years or so," says Mitsolides. "At the same time, we continuously optimize our applications to reduce NFS usage," he adds. Some of the virtual file systems that the bank has investigated include Ibrix Fusion, IBM's General Parallel File System (GPFS) and NetApp's SpinFS.

Eventually, Lehman plans to move a large number of legacy Monte Carlo, Black Shoals-type and other pricing applications off its current production grid and onto its smaller grid, which could take between 12 and 24 months to accomplish, Mitsolides estimates.

To help with the migration, Mitsolides' team is looking forward to the next release of the Symphony platform, which is expected in September and will support multiple versions of the application in the same environment. It should allow you to migrate applications from one version to another at your own pace, explains Mitsolides.

Presently, the test gird is able to handle testing individual trade books, but on the weekend "it borrows CPUs from the large grid automatically so we don't have to manually reallocate resources every time," says Mitsolides.

Lehman officials also expect to eventually link the various grids geographically so that jobs in New York could exploit grid resources in London and Tokyo offices.

"By adjusting the granularity of our jobs, we can take advantage of London or Tokyo hardware-even though the cross-continent network latency is much higher than the latency of the local grid-and still obtain substantial benefits," he says.

Lehman's London office is already testing its grid deployment and the Tokyo office is operating on the grid the bank developed "ages ago," says Mitsolides.

Rob Daly



Click here for a printer friendly version

http://www.dealingwithtechnology.com/public/showPage.html?page=printer_friendly_imd&print=698085