Previous | Table of Contents | Next |
There was no money budgeted, however, for upgrading approximately 329 servers, with only 2M bytes of RAM, to the 4M bytes of RAM that Paradox 4.0 required to operate. Therefore, Borlands solution was cost prohibitive.
Ultimately, the solution was somewhat unsatisfactory but effective. Each site left one Banyan Vines server at 4.11(5) that contained the Paradox NET and SOM files. These files allowed Paradox 3.5 to run normally for all the end users that were attached to Vines 5.52(5) servers. Fortunately, Banyan supports mixed versions of its Vines operating systems, so long as they are within one version of each other (e.g., 4.X and 5.X systems can be mixed and still talk to each other).
The last stage was opening the main Cisco router, which was connected to the RTC and FDIC networks, to full Banyan Vines traffic from both sides. As part of the initial procedure, certain action items were completed before the wave of router table updates went surging through the respective networks.
First, all RTC Vines servers had their communication buffers set to at least 400,000 before the cut over. All CNS servers had to have a primary drive that was larger than 80M bytes (300M bytes were preferred). All network servers were scanned to ensure that no duplicate server names (or server key numbers) were in the two systems. All end users were prohibited from logging on and using the network beginning at 5:00 PM that Friday. The STDA directory services on each server were stopped on Friday morning and were not allowed to rebuild again until Sunday night. Finally, no production servers were trying to get their Streettalk updates across a WAN link that was slower than 56K bytes per second. The STDA was also stopped to prevent any unnecessary WAN traffic while Streettalk was trying to update the internal routing tables of all the production servers.
Full server backups were completed and verified before the process began, to provide an absolute fallback position in the unlikely event that each organization had to return to a preintegration status. The actual process of the integration was that the servers in each network needed to have their internal routing tables updated to know about the additional 300-plus servers on the other network. Because Vines does this dynamically, the staffs primary job was to monitor the health of each server and, in fact, the whole RIC network while the update process unfolded. Once the main Cisco router was allowed to pass Streettalk traffic and updates both ways, it became much like an information tidal wave slamming into each site. At 9:00 PM Friday night, the information updates were rolling toward the location, and server memory utilization statistics were watched to see if any server started swapping information from memory to disk, a sure sign that the server was in danger of being swamped.
By 10:00 PM, the Streettalk routing updates were in full swing, and the available server memory was dropping to levels too low for comfort. Staff stopped all unnecessary file server processes in an effort to free up and make available more server memory for the routing table updates. By 11:00 PM, the servers in the network had stabilized, and available server memory began to climb, a sign that the critical time had passed and that the network was stabilizing.
At 9:00 A.M. Saturday, the staff returned to restart all the services that had been stopped the night before and began testing the services to which the end users would need access on Monday morning, with the exception of STDA. Once all services were tested, and no lingering problems existed, the staff began to deal with the SIDA issue.
It was decided to permit the regularly scheduled rebuilds to go as planned, but the FDIC would not be included in the rebuilds for another 10 days. This would allow Streettalk time to settle before subjecting it to another data blitz from an STDA that would have added another 10,000 names to the existing 7,000-name data base. The final integration of the two STDAs proved to be modest and concluded the entire data integration process rather quietly.
In retrospect, this network integration might have been overplanned, considering how smoothly it went; however, the Vines 5.52(5) server upgrades that blindsided the team on the Paradox issue were underplanned. Data center operations managers looking at integrating similar or even dissimilar networks save time and aggravation if they include field personnel who will be participating in the upgrade/integration process. Bringing in the senior network personnel, training them, and including them in the integration process made all the difference in the merger of the FDIC and RIC Banyan WANs. People who are actually scheduled to do the work who buy in to the project help breed enthusiasm and interest in the success of the project and provide unexpected benefits.
For example, field personnel identified many incidental problems that no one at corporate headquarters had even considered. The success of the overall project was ensured by the corporate people involving the field network engineers.
Every effort should be made to get the hardware and software vendors involved at the beginning of the project to assist and comment during the planning stage. Banyan Systems placed one of its certified Banyan engineers on site at headquarters to help with the integration process. Additionally, Compaq sent its engineers out to RTC sites to help resolve the SystemPro hardware problems that were first encountered when the new drives were installed before the operating system upgrades.
Each vendor that was contacted and included in the upgrade planning before implementation improved the chances for success with either technical advice or on-site involvement. RTC suppliers also contributed in the upgrade process when involved.
Finally, communication among all the players in an upgrade of this size is critical. Information is useless unless it is in the hands of the right people at the right time. Using Banyan E-mail, Lotus Notes, and teleconferencing, headquarters staff and field engineers were able to communicate critical pieces of information to the proper people in a timely manner. As a result, the upgrade and integration success was accomplished with existing personnel who were also responsible for the day-to-day operations of the existing user community.
Previous | Table of Contents | Next |