
Discovering and managing our 10,000-plus nodes appeared to test the upper limit of Spectrum running on a Sun Ultra II with 256 MB of RAM. We say "appeared" because Spectrum reported that it was managing 30,000 objects, but could not tell us how many nodes that represented. Granted, this was without any tuning, which would no doubt improve performance, and moving the GUI off the server would have given the interface more snap. All in all, while running the GUI, the server remained at less than 50 percent utilization and didn't crash, a significant point in light of some of the other products' performances.
Spectrum's GUI desperately needs a redesign. It's full of right clicks and context launching, but they complicate matters. For instance, a simple find or move one layer up the map--a single action for Netview--is a four-step procedure here. And even when executed, some tools are not useful; the database object IDs must be understood to extract data.
Spectrum has a number of views that provide network status from a device perspective. By default, all nodes have statistical displays that report packets in and out, errors, percent packets delivered, forwarded, transmitted, error, discarded, router load, packet rate, percent error and percent discarded, all of which can be viewed by total, delta or accumulation. For example, Cabletron has improved the diagnostic abilities of its port-level views by displaying "Load," MAC (Media Access Control), IP, Ifindex and a mapping of the devices/segments connected to each port. Load is a Cabletron-derived statistic combining CPU utilization, number of packets in and out, and memory utilization, all via a Spectrum SpectroWatch. This can monitor all or specific interfaces for threshold violations.
In conjunction with Spectrum alarm notification, correlation across events is possible but not easy. It's a point-and-click operation to create logical comparisons, which Spectrum doesn't make any clearer since it doesn't provide explanations of the variables and possible interpretations. (Netview does a thorough job of this.) We also found it impossible to compare attributes across different types of devices, even though we needed to do this to create a list of devices that had had problems the previous day. It's a common requirement, but one that Spectrum couldn't fulfill.
But Spectrum did show off its intelligent alarming by immediately transferring SpectroWatch status to the icons; when a SpectroWatch threshold was violated, it was alarmed in the icon, without any special setup.
Spectrum divides reporting functions into three sections--format, generate and view--a setup that promises power, but Cabletron's implementation ended up merely creating extra navigational steps. It has added predefined templates for alarm, inventory, event, availability and statistical information, which makes formatting easier, but choosing the right data requires an understanding of Spectrum's data structure and internal methods of handling third-party infrastructure devices.
We wanted to list the 200 Cisco 1900s on the university network, but we couldn't do it because Cisco had used differing system object identifiers for different 1900 models. Cabletron provided for our tests a beta Cisco 1900 management module (which should have been released by the time you read this) that addressed the differences in the 1900 sysOIDs. Compare this delayed response with Loran's overnight turnaround to resolve similar SNMP inconsistencies, and it points to Spectrum's lack of flexibility.
Also missing in Spectrum is any reporting on network services known through well-known TCP/IP ports, as well as any Web-enabled views of the network.
|