Conference netcad::hub_mgnt

2514.0. "Poor Performance or OverSaturation" by PEEVIE::GREID () Thu Jul 13 1995 13:19

    Here at REO we seem to be experiencing a great deal of response
    problems on two particular DECHub 90 configurations. In both cases
    these Hubs are supporting office thinwire connections via DECrepeater
    90Cs and the Hubs have their own DECagent 90 and Probe 90 running the
    latest software, the Bridges are running V3.1.
    
    I can ping both the Agent 90s and the Probes 90s and can use Hubwatch
    to talk to the Agents but cannot communicate with the Probes as the
    Data Capture window times out with being unable to SNMP to them. Now
    I have tried Probewatch from various different platforms and nothing
    works, even a PC on the same Hub. However the Probes when moved will
    work elsewhere on other subnets in other Hubs.
    
    Now one of the Hubs particularily is showing massive variations in
    utilization figures. I have an HP Analyzer on it and the utilization
    varies from 3-8% to quite frequently shooting up from 10% to as much as
    90% with frequent bouts of being consistantly in the 20s, 30s or even
    40% range. I have found that any IP data transfers between PCs
    particularily CELEBRIS running Windows 95 will push up the utilization
    of the Hub to sometimes mid 50% and remain there until the copy is
    complete, and this is just for one PC copying a 30mg Windows 95 kit
    from and PC server. We have also looked at PCs running Windows 3.1
    copying via IP from a Vax running UCX and the utilization bumps up
    again to mid twenties and even mid thirties until the copy is complete.
    
    Now I realise this is not a PC networking notes conference and I have
    been down the road of swapping hardware eg Bridges, Power Supplies, HUB
    backplanes etc... but are these HUB backplanes man enough for this
    work, or am I expecting too much out of them, ie have I too many file
    transferring, mosiac playing, performance testing users with
    approximately 130 stations for a bob standard HUB 90 to handle?
    
    I really would appreciate any thoughts on this, as I don't understand
    why all of a sudden IP transfers are impacting this so much and what I
    can do to relieve the pressure, as chucking hardware at it seems silly,
    and maybe upgrading the backplane to a HUB900 is a waste of time as its
    still only 10mgs.
    
    Giles 

    You should expect the same level of performance from a DEChub 90 as
    from a piece of ThinWire coax - logically they are the same. That is 
    10Mbs of bandwidth, however you want to use it. (in reality less, due 
    to collisions and interpacket gaps) 
    
    If the servers are not in the same DEChub 90, then you have some delay 
    across one or more bridges. If other segments in the path between the 
    PC and server also go up to 50% or higher, then it is quite probable 
    that the DECbridge 90s may be dropping some packets, causing 
    retransmits from the end nodes, and thus causing the high utilization 
    to be prolonged.
    
    Regarding the probe problem, which version of firmware are you running? 
    Also, are you running V3.2 of PROBEwatch? While it may be that the 
    probe chokes for too long on the high utilization, it also could be 
    a V1.0 problem that has been resolved.
    

    Hi,
    
    	Well both the Probes are running V2.4 and the Probewatch version is 
    V3.2, but the Data Capture Window consistantly fails with not being
    able to SNMP to the Probes.
    
    Giles.

    Further to my last note, both Probes can be pinged ok, however when
    invoking the test probe utility one fails to on the communities, the
    othet works on all the test but still SNMP fails to talk to the Probe
    on a Data Capture.
    
    Giles.

    The discussion regarding the PC Windows 95 IP pounding the network from
    fast PCs is now being discussed in note #582 in the Windows95
    conference.
    
    Giles.