Title: | DECmcc user notes file. Does not replace IPMT. |
Notice: | Use IPMT for problems. Newsletter location in note 6187 |
Moderator: | TAEC::BEROUD |
Created: | Mon Aug 21 1989 |
Last Modified: | Wed Jun 04 1997 |
Last Successful Update: | Fri Jun 06 1997 |
Number of topics: | 6497 |
Total number of notes: | 27359 |
Sandia National Labs, Livermore California is a Beta test site for the GIGAswitch. One of the goals of this Beta test program is to examine the behavior of a GIGAswitch when a serious load is applied. Toward this goal, the following configuration is being used for testing: 1) 1 GIGAswitch with 11 two-port line cards 2) 10 DEC AlphaAXP upgrade systems (AKA DECstation 5000/133) 3) 10 SGI Iris 4000's 4) 10 IBM R6000's 5) 10 SUN Sparc 10's 6) 10 HP workstations (I don't know what model) 7) 1 DECstation 5000 running POLYcenter Network Manager V1.3 8) 9 DEC FDDI concentrators Sandia personnel have written some programs and shell scripts to initiate data transfers between individual systems, a vendor specific pair at a time or all at once.To date, "base rate" tests have been run between pairs of like systems to determine the memory to memory transfer rates each vendor's systems could achieve on an idle FDDI network. There are two modes for this testing. The first mode will cause traffic between individual pairs of systems on an idle FDDI network; the second mode will run between two like systems seperated by the GIGAswitch. A comparison of the two sets of results from these tests (actually 10,000 iterations of each test) should yield a measure of GIGAswitch induced latency. A second set of tests uses a Sandia written program to start the transfer of data through the GIGAswitch between 25 pairs of workstations. In effect, this produces 25 pairs of machines doing memory to memory transfers via the GIGAswitch. Each test runs a little over 70 seconds. The intent here is to study the performance of the GIGAswitch in a high traffic condition. Data from most of these tests has been collected and will be analyzed over the next week or two. As results become available I will add replys to this note. Now for the questions: If you had this configuration and any Polycenter products you wanted, 1. Which products would you configure? 2. What alarms would you use? 3. What information should I trap/export/record/graph... to show how well Polycenter*** will work to monitor high performance networks? Regards, Glen R.
T.R | Title | User | Personal Name | Date | Lines |
---|---|---|---|---|---|
5304.1 | Alarm rules for Gigaswitch | CUJO::HILL | Dan Hill-Net.Mgt.-Customer Resident | Wed Aug 18 1993 01:41 | 30 |
Glen, What can be monitored by the Network manager is largely dependent on what the agent on the gigaswitch is providing. Engineering is working on adding some suggested MIB objects. Please make your requests known to them if you have additional requirements from customers. Here are some examples of MIB variables to watch: dec ema sysobjid bridges GIGAswitch GIGAversion1 gigaBox psc pscStatus dec ema sysobjid bridges GIGAswitch GIGAversion1 gigaBox psc pscBackplaneStatus dec ema sysobjid bridges GIGAswitch GIGAversion1 gigaBox psc cabinetTemperature dec ema sysobjid bridges GIGAswitch GIGAversion1 gigaBox psc rightPowerStatus As far as performance is concerned, there is not much provided by the current agent on the Gigaswitch that will give you an idea of what's going on. You get things like total packets received, but there is no "rate" info available from the agent and DECmcc does not allow you to perform complex alarm expressions, so you're out of luck until the agent is modified. -Dan |