[j-nsp] M10i Version 7.0 R1.5 Per-packet Load Balancing

Harry Reynolds harry at juniper.net
Mon Jun 13 16:53:48 EDT 2005


How long is a piece of string?

The more flows the better, and the more variance per flow the better. I
can look up the specifics on the fields that we hash against, but I
suspect someone has already posted to this list. I know that upper layer
protocol/port can come into play, and this is why a plain ICMP vs a
source-routed ICMP might hash differently despite the same address
pairs.

Regards


PS> Two time the distance from the center to either end.


> -----Original Message-----
> From: Thomas, Steven [mailto:SThomas at birch.com] 
> Sent: Monday, June 13, 2005 1:35 PM
> To: Harry Reynolds; juniper-nsp at puck.nether.net
> Subject: RE: [j-nsp] M10i Version 7.0 R1.5 Per-packet Load Balancing 
> 
> Ok, then how many flows do I need to determine whether a 
> given strategy is working?  I was originally trying to set up 
> some COS based forwarding but I couldn't seem to get that 
> working either so I dropped back to simple load-balancing.
> 
> -----Original Message-----
> From: Harry Reynolds [mailto:harry at juniper.net]
> Sent: Monday, June 13, 2005 3:26 PM
> To: Thomas, Steven; juniper-nsp at puck.nether.net
> Subject: RE: [j-nsp] M10i Version 7.0 R1.5 Per-packet Load Balancing 
> 
> 
> Note that per-packet is really per flow, and that a hack algorithm is
> used to select the next hop. I suspect that the presence of 
> option when
> performing a source-route is affecting the hash, and 
> therefore altering
> the next hop. Put another way, it is not uncommon to see per flow
> balancing not select all possible next-hops when there are only a few
> flows because not every bit in the IP and transport headers is hashed
> against.
> 
> 
> HTHs
> 
>  
> 
> > -----Original Message-----
> > From: juniper-nsp-bounces at puck.nether.net 
> > [mailto:juniper-nsp-bounces at puck.nether.net] On Behalf Of 
> > Thomas, Steven
> > Sent: Monday, June 13, 2005 1:22 PM
> > To: juniper-nsp at puck.nether.net
> > Subject: [j-nsp] M10i Version 7.0 R1.5 Per-packet Load Balancing 
> > 
> > I have been through the documentation and the archives of 
> > this mailing list and I'm at wit's end.  I believe I've set 
> > this up so it should work but it doesn't seem to EXCEPT when 
> > I ping with the route-record option on.  Then it works 
> > perfectly.  Other traffic, such as plain pings, ftp and 
> > telnet always take the same route.  
> > 
> > This is the Junos config:
> > 
> > interfaces {
> >     fe-0/0/0 {
> >         unit 0 {
> >             family inet {
> >                 address 10.10.152.253/24;
> >             }
> >         }
> >     }
> >     fe-0/0/1 {
> >         unit 0 {
> >             family inet {
> >                 address 10.10.153.253/24;
> >             }
> >         }
> >     }
> >     at-0/2/0 {
> >         mtu 9192;
> >         clocking internal;
> >         atm-options {
> >             vpi 15;
> >         }
> >          unit 15102 {
> >             point-to-point;
> >             vci 15.102;   
> >             family inet {
> >                 address 172.16.1.5/30;
> >             }
> >         }
> >     }
> > routing-options {
> >     static {
> >         route 192.168.168.0/24 {
> >             next-hop [ 10.10.153.1 172.16.1.6 ];
> >             metric 1;
> >             preference 150;
> >         }
> >     }
> >     forwarding-table {
> >         export load-balance;
> >     }
> > }
> > policy-options {
> >     policy-statement load-balance {
> >         from {
> >             route-filter 192.168.168.0/24 exact;
> >         }
> >         then {
> >             load-balance per-packet;
> >         }
> >     }
> > }
> > 
> > FE-0/0/0,10.10.152.0/24 is connected to a Linux box which is 
> > my traffic source.  Fe-0/0/1,10.10.153.0/24 connects to 
> > virtual router A on an ERX and at-0/2/0.15102 connects to 
> > virtual router B on the same ERX.
> > 192.168.168.0/24 is an ethernet containing two Sun boxes 
> > which are my traffic destinations and resides on virtual 
> > router B. Virtual router A and B are directly connected with 
> > another ethernet.
> > 
> > This is the output of show route 192.168.168.0/24 extensive:
> > 
> > 192.168.168.0/24 (1 entry, 1 announced)
> > TSI:
> > KRT in-kernel 192.168.168.0/24 -> {10.10.153.1, 172.16.1.6}
> >         *Static Preference: 150
> >                 Next-hop reference count: 2
> >                 Next hop: 10.10.153.1 via fe-0/0/1.0
> >                 Next hop: 172.16.1.6 via at-0/2/0.15102, selected
> >                 State: <Active Int Ext>
> >                 Age: 4:33:55    Metric: 1       Tag: 0
> >                 Task: RT
> >                 Announcement bits (1): 0-KRT
> >                 AS path: I
> > 
> > {master}
> > 
> > Is there something wrong with my config or is there a bug in 
> > this version or what?  Any help would be appreciated.
> > 
> > Steve Thomas
> > 
> > _______________________________________________
> > juniper-nsp mailing list juniper-nsp at puck.nether.net 
> > http://puck.nether.net/mailman/listinfo/juniper-nsp
> > 
> 



More information about the juniper-nsp mailing list