From: "Doug Bissett" <virtualized_ecs_users@2rosenthals.com>
Message-ID: <000.88540a00136c254b.016@telus.net>
To: "Virtualized eCS Users Mailing List" <virtualized_ecs_users@2rosenthals.com>
Date: Sun, 13 Dec 2009 15:34:59 -0700 (MST)
In-Reply-To: <list-1985253@2rosenthals.com>
References: <list-1985253@2rosenthals.com>
Priority: Normal
User-Agent: PMMail/3.07 (os/2; U; Warp 4.5; en-CA; i386; ver 3.07.04.1489)
MIME-Version: 1.0
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
Subject: Re: [Virtualized eCS] TAP driver killing TCP/IP bandwidth

I am replying to my own post, rather than to much more complicated
later posts...

On 2009-12-10, at 14:55:47, Doug Bissett wrote:
>
>On 2009-12-09, at 18:28:29, Lewis G Rosenthal wrote:
>>
>>Hi, Doug...
>
>Hi Lewis...
>
>>On 12/09/09 05:35 pm, Doug Bissett thus wrote :
>>> On 2009-12-09, at 14:01:33, Lewis G Rosenthal wrote:
>>>   
>>> ...snip...
>>>   
>>>> I'll have to fashion myself a CAT-6 crossover to eliminate the 
>>>> switch from the equation.)
>>>>     
>>>
>>> Most modern switches, and NICs, will automatically reverse the wiring,
>>> if they think you might be using the wrong cable. Just try a normal
>>> cable.
>>>
>>>   
>>Indeed, I thought of the auto-sensing circuitry, but don't know if the 
>>T43 or the new Intel chip (still trying to get GenMAC to work with it) - 
>>an 82567LF-2 - do auto-sense. The switch surely does, which is why I no 
>>longer stock crossover cables. :-)
>
>I did my tests with a standard cable (probably CAT5) that is about 5
>feet long. No trouble with not using a crossover cable.

Tried a CAT5e cable. No difference. <more below>

...snip...
>Okay, first some information:
>Test boxes:
> Asus M3A78-EM motherboard, using GENMAC with a Realtek 8168 NIC 
>10EC:8168.
> IBM ThinkPad T43 (1871-W8M), using GENMAC with a Broadcom NIC
>14E4:167D. This machine also has an Intel wireless  8086:4224, which
>was radio off for the testing.
>
>Both systems are running eCS 2.0 Silver, using NETBIOS over TCP/IP. The
>T43 doesn't have the TAP driver installed. The M3A78-EM did not have
>VBOX installed when I started, and the T43 has never had it installed.
>Both systems also have Virtual PC installed, with the bridge driver. I
>never actually ran VBOX on the M3A78-EM, before doing the tests. I am
>using NETIO126.
>
>PROTOCOL.INI on the M3A78-EM, after adding the TAP driver (I forgot to
>copy it before adding it, but that is probably not relevant):
>===========================
...snip - irrelevant...
>========================
>
>Test results:
>=======================
...snip - irrelevant...
>===============================

>The really interesting numbers are in TCP, and NETBIOS, with and
>without the TAP driver installed. Sending seems to be limited to what I
>see with my 100 Mbs hub in the network. With a direct connection, and
>no TAP driver, send and receive seem to be about the same. The
>interesting thing is that receiving doesn't seem to be affected by
>whatever is slowing down sending.
...snip...

>This is somewhat limited testing, but I think that it shows that the
>TAP driver does limit send speed to something near 100 Mbs, even when
>using a gigabit NIC. In different configurations, the results could be
>different.
>
>I never tried wireless, but I will do that, when I find some time to
>play with it again.

No I won't (perhaps later). I did some more testing:

I did get the numbers to improve by removing both TAP$ and Virtual
Switch, from both systems. However, the numbers were still in the:
Packet size 32k bytes:  14486 KByte/s Tx,  14380 KByte/s Rx.
range. (which works out to about 100 Mbs, if I did the math correctly).

I tried the T43 in windows. It shows that the connection is made at 100
Mbs. Connecting the T43 to another windows only machine, using the same
cable, it does connect at 1 Gbs.

I tried NETIO 1.26 between the two windows machines (the T43, and an
Asus M2N-SLI). I didn't try to record the exact numbers, but running
the M2N-SLI as server the T43 client showed roughly:
Packet size 32k bytes:  39000 KByte/s Tx,  79000 KByte/s Rx.
reversing client and server showed roughly:
Packet size 32k bytes:  79000 KByte/s Tx,  39000 KByte/s Rx.
on the M2N-SLI client.

I then tried with windows server on the M2N-SLI, and eCS client on the
T43:
Packet size 32k bytes:  39000 KByte/s Tx,  103900 KByte/s Rx.
Hmmmm..... If the math is right, 103900 KBytes/sec is almost 1 Gbs.

Okay, windows client on the M2N-SLI, and eCS server on the T43:
Packet size 32k bytes:  39000 KByte/s Tx,  78000 KByte/s Rx.

I then tried half a dozen things to try to get the M3A78-EM (10EC:8168)
to go at anything other than 100 Mbs. It just won't do it, so I tried
the driver for the 10EC:8169 NIC. No change. I don't have anything
except eCS 2.0 Silver, or RC6a, to try on the M3A78-EM.

I have run out of time, again., so I will summarize with:

The T43 sending appears to be speed limited at about 39000 KBytes/sec,
whether I use windows XP-Pro, or eCS 2.0 Silver.

The T43, running eCS can receive at almost 1 Gbs, but seems to be
limited to about 79000 KBytes/sec with windows. Note that windows can
apparently send at 1Gbs, so it must be the windows client on the T43
that slows it down.

The M3A78-EM (10EC:8168) runs at 100 Mbs with the default GENMAC
install. I haven't found a way to change that. This I will follow up in
the GENMAC list.

Both the TAP driver, and VSwitch (Virtual PC 5.1) seems to slow things
down somewhat, but I didn't test this today.

Who started this time waster? :-)

-- 
****************************
From the eComStation of Doug Bissett
dougb007 at telus.net
****************************
... For every expert, there is an equal and opposite expert. - Arthur C. Clarke