I'm a CCNA but can't figure out why my Wireless Internet connection is so slow...

Fellow networkers,

I have a Cable modem at home with download speeds of 13 - 15Mb/s. I
love it! Unfortunately, I only measure those speeds on my wired
computers. The wireless ones (I tried 2 laptops using 2 different
PCMCIA cards) only get around 6.5-7 Mb/s when downloading from the
Internet.

NOTE: when transferring a file from my wired computer to the wireless
one, I did get around 15Mb/s. That's slow but it proves that the
wireless computer can receive more than 7Mb/s.

I am using my new Linksys WRT54GS. Both laptops are P4, 512MB RAM,
Win XP SP2. Here is what I tried to eliminate the obvious:

1. Tried both laptops with the same card. Same results, hence not a
computer problem.
2. Tried both laptops with a wired connection. Got 13Mb/s. Again,
not a computer issue. Router is capable.
3. I tried a different wireless card on both laptops. Back to 7Mb/s.
I guess that rules out the wireless adapter.
4. I changed from WPA to WEP and then to no encryption at all. No
improvement.
5. I tried different wireless channels. No luck.
6. I unplugged my only 2.4GHZ cordless phone. Nope. Router is fine.
7. I moved to the room where the router resides. No improvement. I
guess there is no major signal loss.
8. Brought a new laptop from work with an internal Mini-PCI card. Same
speed. Not a PCMCIA limit.
9. Switched back to the old router (Netgear MR814v2). Same old, same
old.

The only thing I am yet to try is using a USB wireless adapter but I
doubt it will show any better speeds.

Is Wireless Internet severely limited? Am I missing something?

Thanks in advance for any advice,
Dotan

 

hath wroth:

Ever consider the possibilty that you can't figure it out because you
have a CCNA? Just a thought.

Cease bragging. I've only got 1.5Mbits/sec.

It should be much faster than that. I just ran a quicky wireless
benchmark test using IPerf just yielded 18Mbits/sec TCP thruput on my
home WRT54GSv4. It's a bit slow because the neighbors are doing
something on my connection via wireless. See:

It would be best if you take the internet out of your benchmarks. If
you have a wired desktop (not wireless) available, setup IPerf server
on it and do your wireless benchmarking locally. That will also take
the router and cable modem out of the picture.

What hardware version (look on serial number tag)?
What firmware version (see status page)?

Cool. You have 2 machines. Do the IPerf server thing with one of
them in wired configuration.

On the wireless laptop, at what wireless speed are you connecting to
the WRT54GS? Your thruput should be about half the connection speed.

From Sherlock Holmes: When we have eliminated the obvious, the
apparent, and those items unworthy of being checked, what remains,
however improbable, is what we have overlooked and have screwed up.

Do you have simultaneous wired and wireless connections running on the
test laptop? The "route print" command will show the current router
table. Look at the "metric" column to see which route has preference.
Most XP laptops will autoswitch between wireless and wired but I've
seen a few utilities that screw things up.

I'll admit that you did a good job of trying to isolate the problem.
The trouble is that you're apparently doing everything with an
internet benchmarking tool. That's fine for later, but makes it
difficult to clearly identify the culprit. Try it without the
internet first.

No, not that severely.

1. Connect BOTH laptops to the WRT54GS via CAT5 cables. Unplug the
cable modem to keep it out of the way. Run IPerf benchmark in TCP
mode (with default parameters). At 100baseT-FDX, you should get about
80Mbits/sec or more. Basically, we're testing the IPerf software and
computers with this test.

On the server:
IPerf -s
On the client:
IPerf -r -c ip_address_of_server
The -r will do a non-simultaneous bi-directional test in case the
problem is asymmetrical.

If the numbers here are unusually low, check:
netstat -s | more
for any IP and TCP layer errors. Any errors are probably at the MAC
layer, but you'll need to fish those out with the ethernet card
diagnostics.

2. Now, replace one of the CAT5 cable with a wireless connection and
run the same IPerf benchmarks again. Note the wireless connection
speed. You should get about half the connection speed.

Let us know what you get here and we'll blunder onward depending on
the wireless benchmark results.

 

Please check the current firmware version of the WRT54GS (take note of
the hardware version to ensure that you selected the firmware version
correctly if there is a need to upgrade it). Go to linksys.com/download
and compare if you have the latest. You can take a look on the Firmware
Release Notes.

If you have the latest firmware, check if you need to update the
drivers for your card....

 

Thank you Jeff and Annie.

I will run IPerf tonight andwill then report the results. I wasn't
aware of it. You're right about the CCNA. If it's not Cisco, they
won't mention it... It's still a good-to-have certification.

For Internet speed tests I used speakeasy.net/speedtest. My WRT54GS is
hardware Ver 6 and I installed the latest firmware from Linksys which
is now 1.51.0.

I usualy get 48-54Mb/s as my connection speed. I thought that my old
Netgear router (.11b) was at fault and according to you 5.5-6Mb/s is
about as much as I could get from it anyway. It was time to get a new
one...

I'll be back later with the results.

Thanks again,
Dotan

 

hath wroth:

The WRT54GS v5 and v6 have problems. See:
<http://www.smallnetbuilder.com/content/view/26843/51/>
In general, the v6 is the same as the v5.
<http://www.linksysinfo.org/forums/showthread.php?t=47124>
I've tried to convince several WRT54G v5 units, that my customers
dragged home, into working, but gave up and exchanged them for other
units.

With an 11Mbit/sec 802.11b connection, the best you can theoretically
do is about 6Mbits/sec. See chart at:

Just a note. Many of the wireless features will actually slow you
down if they are not used. 802.11b compatibility will limit you to
about 15Mbits/sec if there is an 802.11b client connected. Similarly,
if you have the Afterburner or Speedboost enabled, but not in use, it
will reduce the peak speeds to non-SpeedBoost enabled clients. Try
disabling all the fancy features and limit your testing to stock
802.11g only.

 

iperf runs two parts, a server and a client.
On one machine
iperf -s
On the other machine
iperf -c ip_address_of_machine_one
will run a default 10 second transmission from machine one to two.
This can be across OS types, and I run from an internet Unix system through
my NAT router to my PC.

 

DOH !!!

 

Well, there does seem to be some confusion here between cause and
effect. Troubleshooting expertise is not obtained by studying for and
passing exams. It's obtained by getting one's hands dirty and
developing an understanding of how things really work. My theory is
that once one understands how something works, the troubleshooting is
merely the systematic elimination of probable causes[1]. The
inability to get one's hands dirty and get real experience is also why
phone tech support is often useless for troubleshooting.

I have nothing against certifications and certificates. I have a
personal attachment to certificates as I partially supported myself
during college days running a diploma mill. Certificates also saves
some testing effort during the job application and hiring ordeal. It
also guarantees that the applicant knows all the buzzwords from about
3-5 years ago, so we presumably can talk the same language. For
myself, I'm self-certified and too busy to take the exams. If I need
a certificate, I just fire up the certificate factory software, and
produce one on demand. Something like my old warranty card:
<http://802.11junk.com/jeffl/pics/drivel/slides/diploma.html>
There's another on the wall that proclaims that I'm a "Computer
Expert" which entitles me to be arrogant, obnoxious, egotistical,
pontifical, and short tempered to clients and customers. In the 16
odd years that it's been on the wall, only a few people have noticed
and nobody has complained. Yeah, I like certificates.



[1] Once upon a time, I had a lucrative business driving to the
server farms in the middle of the night to perform server and network
troubleshooting. The highlight of one of these trips was watching two
or three CCNE/MCSE holders, busily trying to restore connectivity to a
rack full of servers, when all the lights were off on the switch.
Repair consisted of plugging the switch back in and then wasting an
hour undoing the damage the certificate holders had done trying to
reconfigure the system (because they didn't cover themselves by
keeping a log or making backups). Certificates don't necessary do
much for troubleshooting.

 

I wouldn't let these guys put you down for "tooting your horn". You should
be proud of your accomplishments. That said, they're right in that a piece
of paper doesn't mean that you can troubleshoot anything. Conversely, not
having that certification and/or education doesn't mean that you can't fix
stuff - it's just that most employers require the certification/education.
OTOH, no one can be expected to fix everything on their own - to believe
otherwise is unrealistic. Anyway, I believe your intent was to inform
others that you have substantial knowledge of networking so that they
don't have to "dumb down" their explaination so that you would understand.

 

Fixing the problem is usually easy...finding what the problem is hard.

 

Nope. Both fixing and finding the problem are trivial compared to
assigning the blame. No problem can be solved unless a culprit is
found. In general, it's best to blame someone that is no longer
around or available to defend themselves. Certainly, never blame
anyone that is needed to repair the problem. Interrogation of the
innocent is also a necessary skill. Grilling them for what they
changed, even after they insist that they didn't change anything, is a
fundamental requirement before attempting any serious troubleshooting.
Often, it is difficult to obtain a blame consensus depending on how
the participants are polarized. For example, it is difficult to blame
a Windoze crash on the company Linux faction. Similarly, it is
difficult to blame a hardware failure on a programmer. The real
problem is that there has been little investigation and almost no
publications on this very important aspect of troubleshooting. If
certification classes included more exercises on finger pointing, buck
passing, interrogation, and office politics, troubleshooting will be
greatly facilitated.

 

Certificates don't necessary do

Or as we used to say when Novell's certs started this whole mess:

CNE means Certified, No Experience.

 

I'm back with my IPerf results.

First, let me clarify that as Johnny stated, my whole point about
mentioning the CCNA was not to brag or to start the interesting
discussion re: "Do certified people know what they're doing?". I
merely wanted you to know that I know a thing or two about networking.
I should have mentioned my 7 years in the industry... I'm proud of my
certifications and can't stress enough the importance of having them
when you maintain a career in a competative environment. There's also
nothing like hands-on experience and this is how I got to my original
troubleshooting of the issue. What a beautiful mix!

Back to the results:

Internet was out of the game.
Both computers wired: 92.6 Mbits/sec
1 wireless (server) and 1 wired (client): 19 Mbits/second. The
reported connection speed at that time was 54 Mb/s.

Unfortunately, I didn't have the 2nd wireless card with me so I
couldn't run the test with both computers connected to the access
point.

So if I understand correctly (based on field experience ), that
proves that my router and access points are just fine. Is there an
unusual overhead for packets from the web that are destined to a
wireless client?

I can theoretically call my ISP but they'll most likely blame it on the
Linksys equipment...

Any suggestions would be welcome.

Thanks again,
Dotan

 

Some of us make a lot of money cleaning up
debris left by EEs and Computer Science
majors. The piece of paper is needed
to show what one has done, but then the
rubber hits the road when one has to perform.
Typically, respond to an assignment and
are met by engineers who have been messing
around for maybe a month, and you're out
of there and back on the airplane in an
hour with a lot of money in your pocket.
Awed engineers will ask "who are you!!
Unfortunately, being good at your job
won't get one much status, and engineers
are brought with H1B visas..But, the piece
of paper is still needed, and good to have.
Cheers!!!

 

hath wroth:

I just couldn't a perfect target when presented to me. It's a
constant subject of debate (and irritation) among my few remaining
friends and aquaintences. Also, be advised that all experience and no
theory also has its limitations. One of my early mentors was a self
taught genius (literally). To him, learning anything new came easily
and was obvious, while I stuggled to understand. However, he often
managed to overlook obvious issues that were basic to anyone that had
a formal electronics education. He eventually had to go back to skool
to get a decent basis for what he learned from extensive experience. I
often see this is myself, where I know how to fix or do something, but
have only a limited understanding on how it works under all the
acronyms. If anything, certifications exposes a person to things that
one would normal not care much about (SNA, X.25, IPX/SPX, etc).

If you have 802.11b compatibility enabled, and a connected 802.11b
client, that's about what you would get. Actually, it would be a bit
less. The maximum speed with 802.11b compatibility disabled, and
running in 802.11g mode would be about 24Mbits/sec. You should see
something close to that but your about 12% low. My guess(tm) is that
you have the Afterburner (Speedboost) mode enabled, but are not using
it. Dive into the WRT54GS setting and turn it off. Methinks you
should then see something closer to 24Mbits/sec thruput. You might
also be seeing some intereference. Try other channels (1, 6, or 11)
and see if the speed improves.

Of course, all this begs the question "What thruput speed were you
expecting"? It is possible to have greater than 25Mbit/sec thruput
using proprietary enhancements such as Afterburner. Just purchase a
compatible client card. You already have the correct router. However,
I don't think you'll be too happy with it because the range at

too lazy to find) on the WRT54G indicate that the speed vs range
tradeoff is not too wonderful for Afterburner. This might be worth
reading as a reality cheque:
<http://74.52.135.83/content/view/24861/99/1/1/>
On the other foot, some of the MIMO router are demonstrating really
impressive speed versus range curves. If you really want to go faster
than 20-25Mbits/sec or go through walls reliably, you might consider
one of these.

Assuming the access point doesn't add any additional latency, you're
wireless to wireless thruput should be half again or 12 Mbits/sec
maximum between 802.11g only clients.

Ahem. Based on how closely you're results match theoretical limits, I
would say you're 12% low in speed. Close, but some tweaking might be
useful. However, based on extensive experience, none of which I've
bothered to record or tabulate, I would say your 19Mbits/sec TCP
thruput is typical for a default configured wireless router.

No, and that is a problem. Your previous benchmarks of 7Mbits/sec
need to be explained. The wireless only performance is "good enough".
The wired performance to the internet at 15Mbits/sec is also running
at the best that the cable modem can deliver. Something else is going
on here that is only common to wireless plus internet. Time for some
more testing.

Go unto:
<http://nitro.ucsc.edu>
and try the tests. Look under "more details" and "statistics" and see
if they offer some more clues. If you have a TCP RWIN problem, or
something similar on your client, it will show it. Also, UCSC.EDU is
on the left coast. See the list of other WEB100 servers at the bottom
of the page for something closer.

Also, there is quite a bit of detail that can be extracted from IPerf
run over the internet. However, I have a problem. I know the IP
address of several IPerf servers on the internet, but I don't know if
I'm allowed to disclose them to the GUM (great unwashed masses). I'm
going to play it safe and keep my big mouth shut. However, a Google
search will find a few. No long tests please as IPERF will saturate
the ISP's backhaul.

That depends on the ISP. Ask them to setup an IPerf server so that
you can do your own testing and not bother their support people.
Unfortunately, I can't even convince my own ISP's to do that, so your
chances of success are probably limited. Still, it's worth a try.

Benchark, test, reconfigure, substitute components, borrow a different
router, try same tests at a friends, and whatever you do, make NO
assumptions. Use your own system to learn the limitations and effects
of various parameters. For example, use IPerf to make a graph of
connect speed, thruput, and range. When things are finally deemed
stable, record some benchmarks so that you have a basis for what
constitutes "normal". Extra credit for setting up SNMP and graphing
performance changes over time using MRTG, PRTG, or RRDTool.

 

50% is an often quoted as a rule of thumb for the overhead on wireless
links
so around 20 Mbs is in the right ballpark