[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference irocz::common_brouters

Title:Digital Brouters Conference
Notice:New common-code brouter family: RouteAbout, DECswitch 900
Moderator:MARVIN::HARTLL
Created:Mon Jul 17 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:929
Total number of notes:3736

901.0. "VNswitch Routing Performance?" by SNOFS1::KHOOJEANNIE (Laziness breeds Efficiency) Thu May 15 1997 00:38

    A VNswitch customer here in Australia is using obsolete routers that
    they would like to phase out of their network, and are considering
    either buying Cisco routers now or waiting for the VNswitch IP
    routing upgrades.
    
    They are concerned that the IP routing code will significantly
    degrade the switching performance as it did for the DECswitches.
    
    Do we have measurements on the routing performance of the VNswitches in
    terms of throughput and latency?  I have seen figures of 100,000pps,
    but how realistic is this?
    
    Can we achieve the same throughput whether we route between the
    Ethernets, FDDI etc or whether we use the VNswitch as a "one-armed" router
    across the VNbus?
    
    Thanks                  
    Jeannie
    
T.RTitleUserPersonal
Name
DateLines
901.1Some performance dataIROCZ::REUTHERThu May 15 1997 12:3646
           <<< IROCZ::USER4:[NOTES$LIBRARY]COMMON_BROUTERS.NOTE;1 >>>
                        -< Digital Brouters Conference >-
================================================================================
Note 901.0                VNswitch Routing Performance?               No replies
SNOFS1::KHOOJEANNIE "Laziness breeds Efficiency"     19 lines  14-MAY-1997 20:38
--------------------------------------------------------------------------------
    A VNswitch customer here in Australia is using obsolete routers that
    they would like to phase out of their network, and are considering
    either buying Cisco routers now or waiting for the VNswitch IP
    routing upgrades.
    
    They are concerned that the IP routing code will significantly
    degrade the switching performance as it did for the DECswitches.
    
    >> Normal case switching where the Mac src and dest addresses are 
    >> already learned and that do not involving translation
    >> is done in hardware on the VNswitch.  I would not expect the 
    >> addition of routing to affect this performance.  Bridge learning,
    >> flooding and translation are handled by the same processor that is 
    >> doing routing so this traffic may be affected by heavy loads of
    >> routed traffic.  We have yet to do mixed traffic performance 
    >> testing. 
    
    Do we have measurements on the routing performance of the VNswitches in
    terms of throughput and latency?  I have seen figures of 100,000pps,
    but how realistic is this?

    Can we achieve the same throughput whether we route between the
    Ethernets, FDDI etc or whether we use the VNswitch as a "one-armed" router
    across the VNbus?
    
    >> We have measured slightly greater than 100k in our lab (Bradner
    >> tests) for Eth->Fddi.  I would expect Eth->Eth to be slightly
    >> higher.  If you use a VNswitch as a "one-armed" router across
    >> the VNbus you will incurr a switch/route/switch path.  Since
    >> the route step is the slowest the throughput should approach
    >> that of routing (though the latency will be higher).  At this
    >> point we have not done any latency testing.  Also, if you turn 
    >> on IP Access Controls (IP filtering) the IP routing throughput will be 
    >> reduced by approximately 15%.
    >>
    >> Tom
    
        Thanks                  
    Jeannie
    
901.2ARP-Routing may be of interest.MARVIN::HARTTony Hart, InterNetworking Prod. Eng. GroupMon May 19 1997 13:2415
If you're using the VNSwitch as a "one-armed" router you may want to consider
using a new feature (in V2) called "ARP-Routing" which is essentially proxy-arp
with a couple of tweaks to make it work better in this situation.

Using ARP-routing (which requires some host re-configuration, unless you're
running Microsoft DHCP Server with service pack #2 installed) the host's will
ARP for every address, the router will respond for off-lan addresses, and the
target host for on-lan addresses (even if the two hosts are in different
subnets).  This means that the packets for that connection will be switched
rather than routed and hence get the performance benifit.

There is a white-paper comming soon which will describe this, and of course
it will be in the V2 documentation.

Tony