TCPmag.com
http://tcpmag.com/
http://tcpmag.com/rss
May 1, 2007
Editor: Gladys Rama (grama 1105media.com)
------------------------------------------------------------
------------
THIS ISSUE SPONSORED BY:
- Experience the Ultimate Cisco Classroom!
http://i
nfo.101com.com/default.aspx?id=37979
- Network Traffic Analysis Using Cisco NetFlow: Free Paper
http://in
fo.101com.com/default.asp?id=38136
- Building a Secure Remote Access Network: Live Webcast
http://in
fo.101com.com/default.asp?id=38137
------------------------------------------------------------
------------
IN THIS ISSUE OF TCPmag.com:
1. Q&A: Making Unequal Cost Load Balancing Work
2. What's New on TCPmag.com
3. Interesting Employment in California, Florida
************************************************************
************
SPONSOR: Two World Class Trainers In Every Online
Classroom!
************************************************************
************
Learn from two highly-qualified Cisco certified instructors
and get
the twice the training at half the cost! Special Bring a
Buddy Offer!
Purchase any class and bring a friend for half the price!
Register online at:
http://i
nfo.101com.com/default.aspx?id=37978
************************************************************
************
1. Q&A: Making Unequal Cost Load Balancing Work
Send your toughest Cisco technical questions to editortcpmag.com
with the subject line "Attn: Scott."
Scott,
Does Enhanced IGRP really do unequal cost load balancing?
One of my
co-workers who monitors interface utilization insists that
it does
not actually work.
-- Dan
------------------------------
Dan,
The short answer is: Yes, it does. But the longer answer is:
Your
Mileage May Vary (YMMV). Much of this depends on how you
have
implemented things, and what you actually expect to happen.
So let's start with some basics. Equal cost load balancing
is most
certainly on by default. The "maximum-paths"
command is used to tweak
this basic behavior. The default (as seen by "show ip
protocols") is
to load balance over four equal cost paths. Older IOS
versions have a
maximum of six paths, while newer IOS versions have a
maximum of
16 paths.
The concept of "equal cost" is determined by
looking at the metric of
all received routes. The output of "show ip route"
can obviously see
what paths have been chosen by the routing protocol and
routing table
for use. The output of "show ip eigrp topology"
can be used to look at
all of the paths received (best or not).
This topology table can help us out in seeing all of the
possible
metric values. The "variance" command will then be
used to allow
unequal cost load balancing. The variance indicates a
multiplier.
For example, if your lowest metric (best path) has a cost of
16000,
then a variance setting of 2 will allow any path with
metrics between
16000 and 32000 to be selected as "best path" and
handed to the
routing table.
Now, what happens from here is really the gist of your
question. And
sticking with the standard Cisco answer, it's nothing short
of magic.
There are a couple of things that happen other than simple
math!
First, in order for a path to be a candidate for path
selection here,
that means that the route must come from a Feasible
Successor router.
This has to do with comparing incoming (received) metric
information.
If the received metrics for one path is higher than another,
it's out
of the running. What we are attempting to do here is simply
handle
LOCAL link-based unequal cost load balancing. That's the
"unstated"
technical definition of what we are doing.
Now, assuming that the feasible distance for multiple routes
is OK,
and we do have multiple paths handed to the routing table,
what
happens next is a traffic-sharing calculation. There are two
methods
to this, and the behavior can be altered with the
"traffic-share" command.
The first mode is "balanced." And just like the
name suggests, traffic
is balanced among all links. So if I have one path with a
cost of
16000 and another with a cost of 32000, I will send two
packets down
the lower-cost path for every one packet down the
higher-cost path.
(Well, sort of, but hang on to that idea for a moment.)
The second mode is "min." Here, regardless of the
fact that multiple
paths are in the routing table, only the minimum-cost path
is used
to send traffic over. So, a great question at this point is:
"If
I'm not using the extra path(s), why in the world would I
want them
in my routing table?"
The answer is convergence time. In the event of one link
disappearing,
there is very little recalculation that needs to be done,
and really
no waiting for this to happen, or asking neighbors about
other paths
(active query). The next minimum-cost link is used
immediately.
So, if you have "traffic-share min" inside your
EIGRP configuration,
it's very possible that your co-worker is correct in that
you really
are still using just one link to send traffic over. But
there's
another alternative.
Newer versions of IOS default to using CEF switching. Cisco
Express
Forwarding is a route-caching technique used to speed up
processing
on your router as it moves packets around. CEF, by default,
works on
a per-destination algorithm. So with this in mind (remember
that idea
I told you to hang on to?), your co-worker may still, sort
of, be
right even though "traffic-share balanced" is
used.
By switching on a per-destination basis, it's entirely
possible that
you have a significantly higher amount of traffic for some
destinations than others. Therefore, the actual utilization
your
co-worker sees on one link versus another may simply be an
indication of this switching choice rather than a deficiency
in
the EIGRP algorithms.
Going back to the idea of 2:1 "packet" sending
that EIGRP would
choose: Your route selection and CEF switching selection
interact a
bit. So a 2:1 ratio of "destinations" may be more
appropriate. If
those two destinations just happen to statistically
represent 70
percent of all your traffic, it certainly would seem like
the
load-sharing stuff was all broken!
If you want to TRULY do load balancing and make all of your
numbers
work all the time, look to your interface commands. We can
change
the way the router processes packets and interacts with CEF
switching.
At any outbound interfaces, if you use the "ip
load-sharing
per-packet" command, you can kick things back where the
ratios of
traffic that you see on the interface utilization more
closely match
what you may expect to see. That way, you can tell your
co-worker,
"Of course, it works that way -- you were just reading
things wrong."
Or something like that.
Hope that helps,
-- Scott
Scott Morris, quadruple CCIE, JNCIE and all-around
Uber-Geek, can often
be seen traveling around the world consulting and delivering
CCIE
training. He has recently stepped up as VP of Curriculum
Development
for IPexpert and will oversee a new consulting practice. For
more
information on him check out http://www.ipexpert.com.
Send your questions for this column to editortcpmag.com
with the subject line "Attn: Scott."
Miss a Q&A? Go online to http://tcpmag.com/qanda/
To comment on this Q&A, go to:
http://tcpmag.com/qanda/article.asp?editorialsid=377
************************************************************
************
SPONSOR: Network Traffic Analysis Using Cisco NetFlow
************************************************************
************
Take the Guesswork Out of Network Performance Management!
In this white paper learn to use Cisco NetFlow to minimize
the effort
and cost of gathering network traffic data, while gaining a
granular
understanding of bandwidth usage without the overhead
associated with
hardware probes.
Download it today!
http://in
fo.101com.com/default.asp?id=38136
************************************************************
************
2. What's New on TCPmag.com
NEWS: "Cisco Monitoring Tool Vulnerable to
Attack"
Cisco Systems Inc. last week warned of a no-brainer
vulnerability
in its Cisco Network Services (CNS) NetFlow Collection
Engine (NFC)
which could expose that product to attack.
http://tcpmag.com/news/article.asp?editorialsid=1202
NEWS: "Extreme Takes the Fight to Cisco"
Cisco rival Extreme Networks last week announced a new line
of fix
configuration switches, dubbed the Summit X250 series.
http://tcpmag.com/news/article.asp?editorialsid=1203
RSS FEEDS ON TCPMAG.COM
If you're running an RSS client, then consider signing up
for feeds
from TCPmag.com. You'll automatically be notified when new
content
is posted. Learn more here: http://tcpmag.com/rss/
------------------------------------------------------------
------------
3. Interesting Employment in California, Florida
Job postings courtesy of Monster.com.
CISCO, SOFTWARE ENGINEER
Position Type: Full time
Location: San Jose, Calif.
Salary: Not specified
Experience: Not specified
Desired Education: CCIE
Responsibilities include providing technical support to FA
sites,
implementing and monitoring new FA software/hardware, and
troubleshooting system problems. Must have extensive
knowledge of
Cisco products and LAN/WAN protocols, and have experience
with
scripting and automation.
To learn more, visit:
http://jobview.monster.com/getjob.asp?JobID=56924334
-----------------------------
CITRIX, SOFTWARE TEST ENGINEER
Position Type: Full time
Location: Ft. Lauderdale, Fla.
Salary: Not specified
Experience: Not specified
Desired Education: Bachelor's degree, CCNA, MCSE
Under limited supervision, the software test engineer will
create
software test plans, procedures and standards, as well as
analyze
test results and create documentation. Must have experience
with DNS,
DHCP and software quality assurance.
To learn more, visit:
http://jobview.monster.com/getjob.asp?JobID=56927449
************************************************************
************
SPONSOR: Building a Secure Remote Access Network: Live
Webcast
************************************************************
************
This webcast reviews key considerations for building a
secure remote
access network and the vital capabilities of an SSL-VPN
remote access
solution combined with secure two-factor authentication.
Brought to you by: Redmond magazine and Secure Computing
Register today!
http://in
fo.101com.com/default.asp?id=38137
************************************************************
************
FREE MAGAZINE OFFERS
Subscribe now to our free monthly magazines:
NEW! Redmond Developer News magazine
https://subscribe.1105pubs.com/sub/RW?WP=NEW
FREE&TC=1&PC=MK5
Redmond Channel Partner magazine
https://subscribe.1105pubs.com/sub/RN?WP=NE
WFREE&TC=1&P=OCP01
Redmond magazine
https://subscribe.1105pubs.com/sub/MI?WP=NEWF
REE&TC=1&P=TCP
Sign up for all our related FREE newsletters today.
https://newsletters.1105pubs.com/nl/RMG.do?NL=49
00&PC=TCPNLF
Encourage your peers to excel!
Please forward this newsletter to any IT professional.
************************************************************
************
To learn how you can sponsor a future edition of this
newsletter,
contact Matt Morollo at (508) 532-1418 or
e-mail mmorollo1105media.com
Contact the editorial staff at editortcpmag.com
Newsletter problems: RED1105service.com
TCPmag.com
Redmond Media Group
16261 Laguna Canyon Road, Suite 130
Irvine, CA 92618-3608
Phone 949-265-1520
************************************************************
************
UNSUBSCRIBE OR CHANGE E-MAIL ADDRESS:
https://newsletters.1105pubs.com/nl/RMGf.do?e=nesstosharedlog.com
************************************************************
************
To review our Privacy Policy, visit our Web site at
http://www.1105
media.com/privacy.aspx
Copyright 2007 1105 Media Inc. TCPmag.com News may
only be redistributed in its unedited form. Written
permission
from the editor must be obtained to reprint the information
contained within this newsletter. Contact: editortcpmag.com
|
