Netgear and 108mbps now working...

Since there has been a lot of moaning about the netgear 108 router and adapters
only running at 54 lately, I thought I would mention that they posted a new
driver last week, and it works for me now at 108 link speed. Range is good
(pretty much all over a 6000 sq ft building) with decent signal strength.

Throughput is not awe-inspiring however, although almost all wireless adapter /
AP combinations I have tried have the same problem, particularly with test
code that uses FDX throughput to measure them. Drivers are horribly
inefficient, even with security disabled entirely. Streaming either in or
out does slightly better, but with simultaneous traffic in both directions,
they all seem to fall down badly.

I was able to get 66.64 mbps sustained throughput with a FDX test doing
parallel send/receives on the netgear 108 router and adapter, using a Dell
notebook. In case you're not used to the terms, that means it was sending
about 33.3 mbps and receiving the same amount in parallel, whereas it *should*
have been capable of about 194 FDX. This test was performed with "Excellent"
signal strength, while the notebook was sitting about 3 feet from the AP.

By way of comparison, any decent 100mbit copper NIC is capable of about 180mbps
FDX using the same test software. So, the 108 wireless is actually only
getting about 1/3 of what a 100mbit ethernet hard-wired NIC is capable of on
the exact same test.

Similarly, gigabit ethernet NICs (in servers with 64-bit PCI/PCI-X slots) are
typically capable of around 1800mbps in the same test, showing that the test
software is capable of getting around 90% of the rated ability, regardless of
interface type.

Nevertheless, wireless advertised speeds seem to be hopelessly "ambitious",
and this is the fastest one I've tried to date, now that the new driver
is available.

--
Randy Howard
2reply remove FOOBAR

It is supposed to be full duplex? Or is is half duplex? It makes a HUGE
difference. Essentially, it means the difference of sharing a pipe with two
way traffic or having two pipes that are dedicated to one way traffic. Just
a thought.

Michael

"Randy Howard" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed).. .
>
> Since there has been a lot of moaning about the netgear 108 router and
adapters
> only running at 54 lately, I thought I would mention that they posted a
new
> driver last week, and it works for me now at 108 link speed. Range is
good
> (pretty much all over a 6000 sq ft building) with decent signal strength.
>
> Throughput is not awe-inspiring however, although almost all wireless
adapter /
> AP combinations I have tried have the same problem, particularly with test
> code that uses FDX throughput to measure them. Drivers are horribly
> inefficient, even with security disabled entirely. Streaming either in or
> out does slightly better, but with simultaneous traffic in both
directions,
> they all seem to fall down badly.
>
> I was able to get 66.64 mbps sustained throughput with a FDX test doing
> parallel send/receives on the netgear 108 router and adapter, using a Dell
> notebook. In case you're not used to the terms, that means it was sending
> about 33.3 mbps and receiving the same amount in parallel, whereas it
*should*
> have been capable of about 194 FDX. This test was performed with
"Excellent"
> signal strength, while the notebook was sitting about 3 feet from the AP.
>
> By way of comparison, any decent 100mbit copper NIC is capable of about
180mbps
> FDX using the same test software. So, the 108 wireless is actually only
> getting about 1/3 of what a 100mbit ethernet hard-wired NIC is capable of
on
> the exact same test.
>
> Similarly, gigabit ethernet NICs (in servers with 64-bit PCI/PCI-X slots)
are
> typically capable of around 1800mbps in the same test, showing that the
test
> software is capable of getting around 90% of the rated ability, regardless
of
> interface type.
>
> Nevertheless, wireless advertised speeds seem to be hopelessly
"ambitious",
> and this is the fastest one I've tried to date, now that the new driver
> is available.
>
> --
> Randy Howard
> 2reply remove FOOBAR
>

"Randy Howard" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed).. .
>
> Since there has been a lot of moaning about the netgear 108 router and
adapters
> only running at 54 lately, I thought I would mention that they posted a
new
> driver last week, and it works for me now at 108 link speed. Range is
good
> (pretty much all over a 6000 sq ft building) with decent signal strength.
>
> Throughput is not awe-inspiring however, although almost all wireless
adapter /
> AP combinations I have tried have the same problem, particularly with test
> code that uses FDX throughput to measure them. Drivers are horribly
> inefficient, even with security disabled entirely. Streaming either in or
> out does slightly better, but with simultaneous traffic in both
directions,
> they all seem to fall down badly.
>
> I was able to get 66.64 mbps sustained throughput with a FDX test doing
> parallel send/receives on the netgear 108 router and adapter, using a Dell
> notebook. In case you're not used to the terms, that means it was sending
> about 33.3 mbps and receiving the same amount in parallel, whereas it
*should*
> have been capable of about 194 FDX. This test was performed with
"Excellent"
> signal strength, while the notebook was sitting about 3 feet from the AP.

since the transmit and recieve share the same radio channels - the system is
half duplex.

So, your figures show that it is more or less working as designed - there is
at most 108 Mbps of raw banwidth available (less if interference is causing
some retries)

Also, the overall throughput should be better when most traffic is in 1
direction, as there will less contention / waiting for the medium to come
free.
>
> By way of comparison, any decent 100mbit copper NIC is capable of about
180mbps
> FDX using the same test software. So, the 108 wireless is actually only
> getting about 1/3 of what a 100mbit ethernet hard-wired NIC is capable of
on
> the exact same test.

again - this doesnt compare like with like. Ethernet tested with large
frames has little overhead in full duplex (but a lot depends what you
measure - IP and ethernet level frame calcs will show different results)

In practice there is a lot of overhead on wireless to do with checking for
access, packet lead in, gaps for switching between transmit and recieve and
so on, so you are not going to get useful throughput anywhere near as close
to the raw bandwidth as with ethernet.

>
> Similarly, gigabit ethernet NICs (in servers with 64-bit PCI/PCI-X slots)
are
> typically capable of around 1800mbps in the same test, showing that the
test
> software is capable of getting around 90% of the rated ability, regardless
of
> interface type.
>
> Nevertheless, wireless advertised speeds seem to be hopelessly
"ambitious",
> and this is the fastest one I've tried to date, now that the new driver
> is available.

wireless advertised speeds are the "raw" data rate. The manufacturers
carefully dont explain about the constraints which make for much lower
speeds in practice.

The other issue that is likely to catch you out is that the 108 Mbps
products seem to expect to have all the 11 american assigned channels
available - this isnt realistic in many situations.
>
> --
> Randy Howard
> 2reply remove FOOBAR
--
Regards

Stephen Hope - return address needs fewer xxs

In article <AAsgc.710$(E-Mail Removed)>,
(E-Mail Removed) says...
> > I was able to get 66.64 mbps sustained throughput with a FDX test doing
> > parallel send/receives on the netgear 108 router and adapter, using a Dell
> > notebook. In case you're not used to the terms, that means it was sending
> > about 33.3 mbps and receiving the same amount in parallel, whereas it
> *should*
> > have been capable of about 194 FDX. This test was performed with
> "Excellent"
> > signal strength, while the notebook was sitting about 3 feet from the AP.
>
> since the transmit and recieve share the same radio channels - the system is
> half duplex.

Ack. I hadn't given any thought to that all. The "branding" showing speeds
in the same way as conventional copper and optical nics caught me out. No
wonder the performance is so much worse than those products.

> Also, the overall throughput should be better when most traffic is in 1
> direction, as there will less contention / waiting for the medium to come
> free.

I'll have to retry using HDX tests and see how it holds up in each direction.

> wireless advertised speeds are the "raw" data rate. The manufacturers
> carefully dont explain about the constraints which make for much lower
> speeds in practice.

Exactly... the marketing people are doing their typical little white lie
thing again.

--
Randy Howard
2reply remove FOOBAR