Dave Hewetson, Focus Group Chairman, opened the meeting by welcoming everyone to the Guildhall and thanking Colin Ashcroft, City of London Corporation, for hosting the event.
Dave is Performance Manager for the COTAX project. It was started in January 1998 and was considered sufficiently lengthy and broad-based to justify a project manager post. It involves the transfer of a system from a VME/IDMS/TPMS/COBOL/AM environment to a HP-UX/ORACLE/TUXEDO/COOL-Gen environment. It is a system involving 2000 active users with 400 offices connected. The 'go-live' date is 10th November 1999. A five-day window will be available for the cutover following satisfactory proof of concept during June-August. This proof will be achieved using a 2000 user workload.
Guiding principles for the project are to achieve as much useful measurement as early as possible, measuring at points which will yield most benefit. To assist Dave there is a performance working group which comprises key EDS personnel and representatives from IT partners. This group meets monthly to sustain project momentum.
Dave spoke about the risks anticipated and identified key performance characteristics to be considered. He hopes to be able to provide a progress report at the AXiS conference and report on achievements at a CMUG meeting towards end-Nov/early December.
The claims by Microsoft for NT scalability are not really supported - certainly at NT 4.0. It cannot be regarded as a mainframe system. The product has been driven by the company's internal dynamics, by the need to keep growing the product. But enterprise users need stability. Bill Gates effectively claimed 2 years ago that corporate users could trust all their systems to NT. Today this would still be unwise. Microsoft seem to alternate between clustering and symmetric multi-processing (SMP) as the most effective way to achieve scalability. Today clustering is in favour.
Des reviewed clustering and SMP architectures. Clustering at NT4.0 offers resilience rather than scalability. We will need to see what NT 5.0 has to offer. SMP tests give an impression that moving from a 4-way to an 8-way system gives more substantial improvement up to 6-way but little further is gained beyond 6-way. A report may be found on the Windows NT Magazine web site : http://www.winntmag.com
In terms of memory NT 4.0, a 32-bit OS, supports up to 4 GB of virtual memory of which 2 GB is for users and 2 GB reserved for the system. The Enterprise Edition of NT 4.0 stretches the user allocation to 3 GB.
There is some scalability at the application level, eg Exchange Server 5.5 which removes the 16 GB limit on Message Store and SQL Server v7.0
NT 4.0 lacks tools at the OS level to arbitrate between workloads but, as reported at a previous CMUG meeting, ICL has a Policy Scheduler which, designed for UNIX, will run on NT. A Performance Monitor is available as an application to pick up system metrics but the data is quite raw and some effort is required to develop the application to meet user needs. Again we look to NT 5.0 to see improvements in this area. NT 5.0 instrumentation will enable events, eg process/thread creation and deletion, physical disk I/O and page faults to be recorded. Detailed metrics for process, threads and disks can be derived and there will be tools to analyse bottlenecks.
Any organisation planning to run their enterprise systems on Windows NT needs to consider a number of issues, eg
account management - dealing with Microsoft
support - what effective channels are available
technical manageability - business internal support
other suppliers like Compaq - are they new one-stop suppliers?
NT 4.0 is not really an enterprise system but NT 5.0 should offer major advances. With Beta 3 now in the field there are good prospects for a number of components to be shipped this autumn. The Data Centre Server - ostensibly what most enterprises will need if they choose NT - will be further down the track. This could mean April 2000. It is wise to wait until SP2 is delivered - early 20001?
This second session from Des looked at Windows 2000 (NT 5.0) He reviewed the shortcomings of NT 4.0 from the Performance Management perspective, ie all tools are snapshot based; there is no accounting function as with UNIX and mainframe systems; per process I/O counters though present in NT 3.1 had disappeared by NT 4.0 The system is good for real-time monitoring though data logging is unsophisticated.
Windows 2000 offers a System Monitor and Event Trace (described as additional data collection technology)
System Monitor (SM) is an embeddable and programmable component which replaces the NT 4.0 Performance Monitor whilst retaining the latter's look and feel. It offers performance logs and an alert service. SM interfaces, eg methods, properties, events, can be accessed by Visual Basic programs for analysis. New system metrics include Plug and Play (PnP) Disk Perf and Disk counters.
What is Event Tracing? It is a recorded and ordered set of events providing information to supplement the standard counters, eg disk I/O, TCP/IP traffic. thread creation/deletion, file I/O. When a trace is run an event log file is created on the PerfLogs directory (.etl extension). Whilst Microsoft talk about detailed analysis tools, as at Beta 3 there is nothing available to look at .etl files. So at present it sounds like DIY!
Comparing SM and Event Trace Data: SM counters have very low overhead and are better for continuous monitoring. There are problems relating system to user. Event Trace carries a larger overhead but is better for detailed analysis or capacity planning. It is good for relating system to user. Microsoft Help actually warns about the overheads, volumes of data, generated by trace logging of file I/O. Caveat User!
No CMUG meeting is considered complete these days without wise words from John Popplewell. This meeting was nicely rounded off by John presenting a look at List_Usage and Elapsed_Time_Monitor. He identified which resources could be monitored and explained why and how the tools should be used.
Output from the tools can be passed to a Journal. It is considered helpful to pass it in CSV format to either the job journal or a public journal. The CSV format is suitable for processing by Excel. ETM is separately licensed by ICL so if you are interested please talk to your ICL Account Manager.
Dave Hewetson thanked the City of London Corporation through Colin Ashcroft for hosting the meeting and providing an excellent lunch; and Des Atkinson and John Popplewell for their presentations.
7. Next meeting
The next meeting is expected to be held late November/early December 1999. Would any member volunteer to host the meeting? Provision of lunch is very welcome but not a prerequisite. Please call Dave Hewetson (Tel no in attendance list at the head of these minutes) if you wish to host the meeting and also if you have a topic(s) which you would like to air or to be aired.
to AXiS Home Page.Copyright © AXiS 1999 Created: 4th June 1999