Just to add if am allowed too.ET (hope you ment Emerging Technologies) is a device that you will add on your network and will Forget.
You can go for  a software version that will need you to patch the kernel of your choice or go for an already made Rack mount box with failover/redundancy capabilities plus Gigabit ethernet.

Ours has't gone down for 3 years.Got a newer one recently at roughly $5000 worth it;-) .Note; It's not just a bandwidth manager (router,firewall,brigde name it) all on Linux! Good for a network tool/machine
Rgrds
Ronny


Mark Tinka wrote:
On Tuesday 19 April 2005 17:55, Guido Sohne wrote:
  
Apologies for the delay in responding. A storm collapsed the mast
serving my area with wireless connectivity.
    

No worries... another redundant connectivity?

  
Last mile is wireless. Motorola Canopy, I believe.
    

What kind of (bandwidth) management capabilities do you get on this device. 
I've heard popular deployment of these units, but haven't had personal 
experience with them yet.

  
I'll look into this. We had to create a private network to route between
the existing internal network and the public network.
    

Hmmh... can you draw and ASCII? Are public/private networks synonymous with 
public/private IP's, or simply network segmentation?

  
Both of these 
networks use public IP addresses and my problem was how to send packets
to the router.
    

>From where? Not sure I understand.

  
I think it should be possible to forward packets from the 
bandwidth manager (if it is setup as a bridge) to a new default gateway.
    

This is the thing - with the bandwidth manager running as a bridge, no 
forwarding is required. Packets are presented to the bandwidth manager 
ingress/egress ports (depending on direction of packet travel), processed by 
the bandwidth management process, transported over the high speed backplane 
to the exit port, and that's it.

You could say it's a switch that analyzes packets, but doesn't route them. Any 
IP addressing information (IP/Mask/Gateway) configured on this device is 
purely for management purposes (remote access, monitoring, e.t.c.), and not 
for production purposes (bandwidth management).

  
I wonder if it is simpler to do that or to keep the existing private
network ...
    

Whichever is simpler and doesn't introduce unnecessary overhead, complexity or 
points of failure.

  
Both. I have settled on Shorewall and I think it is very nice because it
allows you to think at a higher level than iptables rules.
    

Never used this before, but a quick browse on www.shorewall.net reveals it 
provides a user-friendly interface to IPTables.

  
Yes. The box talks directly to the router and proxies HTTP for internal
clients.
    

So the redirection to the cache box is done by the bandwidth manager? Just 
need to make sure I understand this right.

  
Cache hits are returned at full speed.
    

At full speed to the customer, at contracted bandwidth to the customer, all 
the way to the last mile?

  
Once in a while I get  
about 30k/second when one of my requests hits the cache. The cache is
setup to keep many small objects in memory and to keep many large
objects on disk.
    

A very good design.

  
I am recommending to the client that they firewall connections from the
client side as well. The idea is to insulate customers from each other
and to avoid the client's internal network being a 'soft' spot, security
wise. Do you have any recommendations for this?
    

Being a wireless network, it's definitely a broadcast multi-access medium. So 
if you put customers in a single subnet, e.g., /28, those customers will more 
than likely *see* each other.

I usually like to simulate a point-to-point leased line connection on a BMA 
network by using Ethernet sub-interfaces (not meant for this, but does the 
trick). This would mean using 802.1q VLAN trunking on your switch/router 
ports. You then have an Ethernet sub-interface for each customer, assigning 
a /30 for the point-to-point connection - this would be a first-line filter 
of broadcast messages.

You need to be wary of router performance when running 802.1q trunking.

  
Acknowledgement packets and other TCP connection overhead that involves
very small packets are sent first. It takes up less than 4kbit / second
but enables other incoming connections which are waiting for an ACK to
send the next packets.

HTTP traffic, mail traffic etc are sent next but have access to much
higher bandwidth. What happens is that there is a slight pause while
waiting to send the overhead packets, and everything else then comes in
faster as a result.
    

Sounds like QoS.

  
We also cache DNS traffic since we realized that the 
customer's web browser first makes a DNS lookup, then the transparent
proxy makes the same DNS lookup.
    

This is pretty standard.

  
There is no internal DNS, so caching 
makes things respond faster by avoiding the satellite uplink/downlink
latency where possible.
    

Yes. 


  
Do you have any recommendation for bandwidth managers? Thanks for taking
the time to respond. It has been very interesting indeed.
    

Well, aside from the (lack of sufficient) support, I do have very good 
experience with ET. The device has really great price/performance 
characteristics, when pitched up against the 3 other big names - Allot, 
Packeteer and Xedia. Overall, you can easily have 20 times more throughput on 
an ET that costs up to 4x less than the other vendors.

It runs a proprietary bandwidth management program on (a tuned) FreeBSD or 
Linux (prefer to boot FreeBSD), and will push 100BaseT as well as Gig-E.

If you let me know what your requirements are, I could give you a specific box 
to look at, and you can take it from there.

Mark.

  
-- G.
_______________________________________________
LUG mailing list
[email protected]
http://kym.net/mailman/listinfo/lug
%LUG is generously hosted by INFOCOM http://www.infocom.co.ug/
    
_______________________________________________
LUG mailing list
[email protected]
http://kym.net/mailman/listinfo/lug
%LUG is generously hosted by INFOCOM http://www.infocom.co.ug/

  


-- 
***************************************************************************
  / ''We can't become what we need to be by remaining what we are''\
  \ ,,                                                           ,,/
***************************************************************************
_______________________________________________
LUG mailing list
[email protected]
http://kym.net/mailman/listinfo/lug
%LUG is generously hosted by INFOCOM http://www.infocom.co.ug/

Reply via email to