🚀 go-pugleaf

In article <35187EC3.E75031D8@vickers.com.hk>,
Kimmy Chan  <kimmy_chan@vickers.com.hk> wrote:

>However, our concern is the spanning tree protocol which incurs a
>switch-over time (vary from 5sec to 50sec) that will make all
>connected IP sessions disconnect and they need to be established
>again.

5 seconds of lack of connectivity isn't long enough for TCP to notice.
In my experience, 50 second isn't long enough either.  Now, when you
hit 5 minutes you have to start praying hard.

Kimmy Chan wrote:

> Hi,
>
>     My company's business is highly dependent on network service for
> both internal LAN access and external data feeds. Currently, we don't
> have any redundancy build in any network path to backup the system and
> we are planning to do so. We are using Cisco products (switches, routers
> etc) and some network vendors proposed Fast Ether Channel with spanning
> tree protocol to provide alternate network path. However, our concern is
> the spanning tree protocol which incurs a switch-over time (vary from
> 5sec to 50sec) that will make all connected IP sessions disconnect and
> they need to be established again. Any other better methods avoid the
> session disconnect?
>

I agree with other people who have posted that TCP should be able to
withstand pauses in the low tens of seconds.  More transparent options
involve sophisticated NAT.  See:

 "Load Sharing using IP Network Address Translation (LSNAT)", Der-Hwa Gan,
  Pyda Srisuresh, 03/09/1998,
ftp://ftp.ietf.org/internet-drafts/draft-srisuresh-lsnat-02.txt

Kimmy Chan wrote:

> Hi,
>
>     My company's business is highly dependent on network service for
> both internal LAN access and external data feeds. Currently, we don't
> have any redundancy build in any network path to backup the system and
> we are planning to do so. We are using Cisco products (switches, routers
> etc) and some network vendors proposed Fast Ether Channel with spanning
> tree protocol to provide alternate network path. However, our concern is
> the spanning tree protocol which incurs a switch-over time (vary from
> 5sec to 50sec) that will make all connected IP sessions disconnect and
> they need to be established again. Any other better methods avoid the
> session disconnect?
>
> Any comment and suggestion is welcome.
>

You may want to explore some load-sharing network-and-port translation
architectures, such as
ftp://ftp.iepg.org/internet-drafts/draft-srisuresh-lsnat-02.txt.   Also go
to the www.nanog.org home page and find Ed Kern's presentation in the
Albuquerque meeting notes.

Without knowing more about the application, it's hard to suggest
solutions.  Multihoming is a complex topic, and the first requirement is to
define the problem.  Your best solution may not be routing based, but
involve DNS, NAT, etc.  See
ftp://ftp.iepg.org/internet-drafts/draft-berkowitz-multirqmt-01.txt

If you have a very well-defined and controlled network, you could adjust
the spanning tree timeout parameters in most Cisco equipment to reduce the
cutover time.

Of course, this could be an excellent lessen in how to crash a network. 
Allow plenty of time for analysis & experimentation, or find an expert in
STP.

-- Rex

In article <35187EC3.E75031D8@vickers.com.hk>, Kimmy Chan
<kimmy_chan@vickers.com.hk> wrote:
>    My company's business is highly dependent on network service for
>both internal LAN access and external data feeds. Currently, we don't
>have any redundancy build in any network path to backup the system and
>we are planning to do so. We are using Cisco products (switches, routers
>etc) and some network vendors proposed Fast Ether Channel with spanning
>tree protocol to provide alternate network path. However, our concern is
>the spanning tree protocol which incurs a switch-over time (vary from
>5sec to 50sec) that will make all connected IP sessions disconnect and
>they need to be established again. Any other better methods avoid the
>session disconnect?

Kimmy,

Fast Ether Channel, as far as I know, provides more bandwidth and achieves
some redundancy.  However, if the module dies; you lose connectivity.  It is
a good network resilience suggestion.  If Fast Ether Channel can span
through different modules or box, it is more safe.

Cisco router has got a HSRP (Hot Standby Routing Protocol) feature to do
switch over upon detecting an interface failure.  I don't have switching
time figures on hand.  But I believe it is quicker than IEEE802.1d Spanning
tree.

Some vendor, say 3Com, they implement backup link on their switches.  It is
proprietary, non-Spanning tree compliant.  Switching is in the region of 100
milli-seconds and below.  This solves your problem.  Give them a ring and
find out.

On host side you may consider a fault-tolerant transceiver.  Chipcom (now
3Com) and Madge has got a lot of these products for 10Mbps ethernet.  I
haven't seen 100Mbps product yet.  May be I am not up-to-date.  Ask the
vendors for ythe 100Mbps fault tolerant tranceiver availability.  It makes
sense to have 100Mbps on server and host side because it is the bottle-neck.

Hope this help.

Regards,
Roger Luk

Kimmy Chan wrote:

> Hi,
>
>     My company's business is highly dependent on network service for
> both internal LAN access and external data feeds. Currently, we don't
> have any redundancy build in any network path to backup the system and
> we are planning to do so. We are using Cisco products (switches, routers
> etc) and some network vendors proposed Fast Ether Channel with spanning
> tree protocol to provide alternate network path. However, our concern is
> the spanning tree protocol which incurs a switch-over time (vary from
> 5sec to 50sec) that will make all connected IP sessions disconnect and
> they need to be established again. Any other better methods avoid the
> session disconnect?
>
> Any comment and suggestion is welcome.
>
> Thanks,
>
> Kimmy.

Why not use cisco's HSRP (Hot Standby Routing Protocol)?
Faster cutover, easy config (one virtual address for two interfaces)?

Rex Sanders wrote:
> 
> If you have a very well-defined and controlled network, you could adjust
> the spanning tree timeout parameters in most Cisco equipment to reduce the
> cutover time.
> 
> Of course, this could be an excellent lessen in how to crash a network.
> Allow plenty of time for analysis & experimentation, or find an expert in
> STP.
> 
> -- Rex
> 
> In article <35187EC3.E75031D8@vickers.com.hk>, Kimmy Chan
> <kimmy_chan@vickers.com.hk> wrote:
> >    My company's business is highly dependent on network service for
> >both internal LAN access and external data feeds. Currently, we don't
> >have any redundancy build in any network path to backup the system and
> >we are planning to do so. We are using Cisco products (switches, routers
> >etc) and some network vendors proposed Fast Ether Channel with spanning
> >tree protocol to provide alternate network path. However, our concern is
> >the spanning tree protocol which incurs a switch-over time (vary from
> >5sec to 50sec) that will make all connected IP sessions disconnect and
> >they need to be established again. Any other better methods avoid the
> >session disconnect?