Internet 10 times slower in linux than in windows
 

Hi,

I have a home network with a cable modem. And there is three machines: two mahchines are Windows 2000, and the new one is a red hat linux 9.0.

The 3 machines are hooked with a switch. I have a program to fetch web pages from internet. All of the machines have PCI 10/100 enthernet card. My linux box CPU and memory are far better than the Windows 2000 machines. However, the internet in linux is 10 times slower .

This is to my surprise and I found somebody had similar problem with me. But did not find an answer to this.

Thanks

Some things to try:

1. Check MTU size settings for each interface, on Windows check out
http://www.winguides.com/registry/display.php/280/
for Linux, use ifconfig to check and set MTU and other card settings
in /etc/sysconfig/network-scripts/ifcfg-<intf name>

MTU should be the same for all 3 machines.

2. Check the switch settings, try turning off auto-sensing, also see what the speed setting is on each card 10 or 100

3. Check /etc/resolv.conf and make sure that the nameserver entry(s) are correct (correspond to same IP addresses for DNS servers used on windows machines).

If still having slowness problems then check ifconfig again and see if there are any RX or TX errors or collisions. If so, check ethernet cabling to the Linux box.

I don't use any 10/100 hardware, just have read a lot of articles on issues with autosensing on switches.

Might be a silly question -- but how do you access the Internet on the Linuxbox -- is it hooked up to a hub or do you have to go via another machine to get to the internet -- for example is your Windows Box acting as a router or gateway for the Linux box.

Try pinging from the machines --- There's nothing inherently slow about Internet access in Linux or even 'Doze.

What Browser are you using -- Konqueror might be slow if your linux box has little RAM and an old CPU as KDE overhead is quite high.

Try Galeon or the latest incarnation of Mozilla -- it's called redfox / redwolf or something eaually forgettable.

Unnplug the whole kaboodle and just try the linux box alone -- also check you haven't got so many firewalls / virus scanners -- spybots or whatever running that the CPU never gets a look in.

The latest distros of Linux make 'Doze really 'DOZE -ZZZZZZZZH

Finally are all the ethernet cards operating at 100 Mbs -- check because you might find for some reason the Linux connection is only operating at the slower 10 mps.

(BTW this is the Wireless networking -- but same is true if you had wireless cards -- check the connection speed of each).

You might also see if you have the right kernel -- say x686 instead of the older x386 one's.

I can't believe that Linux per se gives slower Internet Access than 'Doze.

thanks for the message.

I am not the first one to find the issue. I am googling the problem and found several people asked this type of question. Some people have a dual boot up and same hardward. And the linux internet is slower than Windows. Bill Gates maybe be love this story! But no good answer yet.

Just to clarify the speed comparison. I am not using any browser to measure the speed. I have a program to download the same web site using the different machines. It took 43 s in windows and 384 s in linux.

The machines connect to the same cable modem. I checked the ethernet card in linux. It says 100MPS during boot time. My switch is netgear RP614. There is no way to turn of autosensing.

Here is the message of ethernet card during boot:
eth0: RealTek RTL8139 Fast Ethernet at 0xd00ae000, 00:0c:76:3c:01:3f, IRQ 11
eth0: Identified 8139 chip type 'RTL-8139C'
ip_tables: (C) 2000-2002 Netfilter core team
eth0: Setting 100mbps full-duplex based on auto-negotiated partner ability 45e1.

If I do ifconfig, here is the message:
eth0 Link encap:Ethernet HWaddr 00:0C:76:3C:01:3F
inet addr:192.168.0.4 Bcast:192.168.0.255 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:17581 errors:0 dropped:0 overruns:0 frame:0
TX packets:14563 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:100
RX bytes:16886281 (16.1 Mb) TX bytes:1884774 (1.7 Mb)
Interrupt:11 Base address:0xe000

lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
UP LOOPBACK RUNNING MTU:16436 Metric:1
RX packets:233403 errors:0 dropped:0 overruns:0 frame:0
TX packets:233403 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:15959656 (15.2 Mb) TX bytes:15959656 (15.2 Mb)


Nothing seems to be wrong. I checked /etc/resolv.conf as well. It is the same as Windows.

The ISP is comcast. Maybe they have windows server, which is optimal for performance on windows client.

I hope that I could find a way around this, or I have to stick to Windows. What a disappointment!

Interesting problem, see if there is similar difference in speed susing different applications (ftp, scp, etc) on different external hosts.

When you mentioned the two times 43 and 384 seconds, for what size of download was that?

scp (secure (ssh) copy) is nice in that it has a progress meter to see if the download is constant or not.

Always check /var/log/messages as always and perhaps dmesg to see if anything strange with ethernet driver on boot up.

----
If still no clues on the problem then you can use Ethereal
http://www.ethereal.com/

to sniff the packets between your linux box and the host downloading from.
You can also run Ethereal on windows and get comparison on the patterns of packets being sent between your local machine and the remote host.

You should be able to get the timings and packet data using Ethereal.

I wouldn't mind helping out how to run ethereal or read the logging output.

try this simple task -- USE FTP on both machines to download a standard file of known size -- ensure the server is not too busy -- so download TWICE or THREE TIMES so you get consistent results.

For windows use WS_FTP or another decent FTP program -- it should tell you your downloading speed and progress.

For Linux standard FTP -- don't use KBEAR or anything like that -- not that there is anything wrong with it it's just by using FTP with as little overheads as possible we can see how thr REAL apps are working for speed without a lot of sophisticated measurement needed.

Using FTP avoids a lot of the browser overhead so should take out the Internet Explorer (Windows) or Mozilla / Konqueror etc (Linux) overhead.

Ensure that these tests are done INDIVIDUALLY as sometimes especially if you go through some type of routers if the machines are working together the position of each on the network can make a difference -- I'm not a network guy so I can't give you a technical answer -- but I've sat behind enough computer screens to KNOW it's true.

If there' s still a difference physically swap the positions of the machines on the network and if necessary swap the cards as well.

If there's STILL a difference -- then your linux has something wrong with it -- try recompiling for optimized kernel, look at boot and start up options -- for example 99% of Linux installations have FAR TOO MANY SERVICES running at boot.

Good Luck and please post when you've found the answer -- I'm curious as my Linux works better than my Doze box.