Requires Free Membership to View
in 2008, but then the market crashed, and
everything was put on hold. So much so, that the second part of the
project was resumed only in 2010.According to Atul Bansal, the assistant vice president of IT for Stewart & Mackertich Wealth Management, a pre-implementation study as well as an evaluation was conducted at that point of time by SMIFS for virtualization of its servers. "We were at the stage of finalizing the vendor for virtualization of servers when the market came crashing down. However, the economic downturn did give us a tremendous cost benefit for server virtualization," recalls Bansal. How? VMware's licensing costs spiraled downward not only due to the market crash, but also due to competition from Microsoft's Hyper-V server virtualization platform, which had entered the market with a free hypervisor.
However, before going for in for VMware, SMIFS had also evaluated server virtualization solutions from Citrix and Microsoft. Bansal feels that it's a myth that Microsoft costs less. "After calculating the licensing, Hyper-V was costing more or less the same as VMware. The management software is being charged separately in Hyper-V. Also, it is not as mature as VMware in terms of support of applications and OS for virtualization of servers," says Bansal. Certain aspects like high availability are not being supported in Hyper-V, but are charged in the components. What's more, there is no fault-tolerance feature available. Citrix also costs almost the same as VMware when it comes to solutions for virtualization of servers. The free version does not support features like VMotion. Thus, SMIFS decided to zero in on VMware. They have opted for the VMware ESXi version.
| |||||||||||||||||
Fault-tolerance has not been incorporated in the licensing model opted for by SMIFS. It includes six cores per processor. The mail server, domain server and back office server are residing on VMs. 90% of the servers are on the Windows platform since NSE and BSE support Windows. The mail server will soon be shifted to the Linux-based platform. Out of the three virtualized servers, the third server is running on the free version of VMware, and it has four VMs. It is running the Direct Market Access application; this application is being run on the Linux platform, which is claimed to be a first of its kind.
Virtualization of servers took around three months. The total cost of the project, including licensing and hardware, was under Rs 15 lakh, with Network Techlab as the implementation partner.
For SMIFS, one of the major doubts was in terms of how successfully ODIN—the online trading application for multiple exchanges—would run on a virtualized server platform. ODIN is a broadcast-based application. Stewart & Mackertich gets a broadcast from different exchanges, and the same broadcast should be available to users on a real-time basis. SMIFS did not get support from Financial Technologies (which had designed this application), because they had never tested it over a virtual server platform. At this point, VMware felt that there should not be any problem since all the applications ran on Windows and Microsoft SQL and supported IP-based broadcast (which is not proprietary). The implementation went smoothly without any challenges. However, during the post-implementation phase SMIFS faced a problem with the ODIN application—as soon as the broadcast started, it would hang. After speaking to VMware and fine-tuning the configuration, things were back on track.
According to Bansal, the major advantage derived from virtualization of servers has been in terms of the ease of manageability. Speaking on the other essential part of the virtualization of servers, Bansal advises, "One should go in for storage consolidation along with virtualization of servers. Otherwise, there are chances that you will miss out on features in thin-provisioning. Also, you are saving a lot on the backup windows because you can allocate it according to your needs."
SMIFS has gone in for an EMC SAN, which has been replicated in its Calcutta office (which also doubles as the DR site). The main objective of this setup is to ensure availability of virtual machines. Instead of normal DR, where one matches hardware to hardware and application to application, in this case SAN to SAN replication is performed, which is offline in nature. This is configured every night, and a snapshot is taken every hour as part of the regular backups. "We are particular that a virtual machine sprawl should not happen, hence we are going about things in a much planned manner. Thus, right now we are creating DR only for those applications which are really needed," adds Bansal.