Linux still surfs slower than Windows

Hi:

I have two Suse 9.1 Linuxes and one Win2k on a LAN behind a Linksys
WRT54G router connected to SBC/Yahoo DSL. The router gets a dynamic IP
via DHCP, and connects via PPPoE. It also receives DNS server
addresses. The LAN hosts have static IPs on subnet 192.168.1.0 and
statically configured DNS server addresses set to the values shown by
the router. These DNS addresses are always the same: 63.203.35.55
and 206.13.28.12 .

The Windows host also has a static IP and manually configured DNS
servers. Also, there is a Win2k VMware machine on each Linux box with
bridged networking and static IP and DNS configurations.

So a total of two Linux boxes, one real Win2k, and two virtual Win2k.

The problem is simply that the Linux boxes using Mozilla or Konqueror
web browsers, surf the web ridiculously slow. Typically 45 seconds to
load a page like www.cnn.com, and 10-15 seconds to load simpler pages
like www.google.com.

The Win2k, both the real and virtual machines, all surf at instantaneous
speed using either Mozilla or IE.

I changed the Linux DNS server configurations to free servers:
205.166.226.38 and 69.67.108.10, which improved speed significantly.
Now Linux surfs about 3-4 times slower than Windows on average for all
web sites. A big improvement, but still unacceptable. Typical 6-8
second loads for www.cnn.com, vs. 2-3 seconds for Windows.


I posted before "Terrible Web Surfing Speed" and will summarize the
results of the suggestions and other attempts at fixing this:

1. The only suggestion which improved matters was to use in the file
/etc/resolv.conf:

options timeout:1

This improved the surfing speed using SBC/Yahoo DNS servers to about the
same speed as using the free servers, still about 3-4 times slower than
Win2k. This is the condition I am in at this time.

2. Switching to DHCP IP and DNS assignments for the LAN clients did not
help.

3. Turning off IPv6 did not help.

4. Captain Dondo suggested: "Well, perhaps your router/ISP is issuing
ICMP redirects and your linux box ain't set up to accept them? I don't
know if Windows accepts redirects by default; it makes it easier for the
user but opens up security holes.... Hmmm. Which way would Windows
lean? As root, run this command:
for f in /proc/sys/net/ipv4/conf/*/accept_redirects; do echo 1 > $f; done

Well I have all these already set to 1 so that can't be the problem:

> for f in /proc/sys/net/ipv4/conf/*/accept_redirects; do cat $f; done
1
1
1
1
1

5. Use the WRT54G as the DNS server. I tried this, and it works. Some
performance is a little bit better than either the free servers or the
options timeout:1, but still about 2 times slower than Windows.

Hmm.

Very close, but still disappointingly slower than Windows. It seems
with the router as DNS, that initial loads of new pages are slow, but
then loading again is very fast. Before, it was the same slow speed all
the time. But Windows still manages twice the initial load speed for
all pages.


Any further suggestions how to improve things would be appreciated.

Good day!




--
_____________________
Christopher R. Carlen
(E-Mail Removed)
SuSE 9.1 Linux 2.6.5

On Fri, 18 Feb 2005 19:40:08 -0800, Chris Carlen
<(E-Mail Removed)> wrote:
>
> Any further suggestions how to improve things would be appreciated.
>
Increase the size of the browser cache?

Bill Marcum wrote:
> On Fri, 18 Feb 2005 19:40:08 -0800, Chris Carlen
> <(E-Mail Removed)> wrote:
>
>>Any further suggestions how to improve things would be appreciated.
>>
>
> Increase the size of the browser cache?


No. That wouldn't explain why the Linux browser chokes on any given
page when the cache is cleared, while the Win2k browser flies when the
cache is also cleared.

I am convinced there is a DNS lookup problem, where the SBC servers
don't work smoothly with Linux requests. The technical details of why
this could be so are beyond my knowledge. I would welcome anyone who
could explain this, as it would seem incredibly important if such a
discrepancy is possible, for the community to have collective awareness
of it and how to affect workarounds.


I have optimized my surfing speed experience through other workarounds,
but none of which resolve the fundamental problem, which persists if I
leave the browsers in their default configurations, and both OSes wset
up with the same DNS configurations.

I still wish I could resolve this fundamental problem.



Good day!


--
_____________________
Christopher R. Carlen
(E-Mail Removed)
SuSE 9.1 Linux 2.6.5

On Sun, 20 Feb 2005 17:15:22 -0800, Chris Carlen wrote:

> Bill Marcum wrote:
>> On Fri, 18 Feb 2005 19:40:08 -0800, Chris Carlen
>> <(E-Mail Removed)> wrote:
>>
>>>Any further suggestions how to improve things would be appreciated.
>>>
>>
>> Increase the size of the browser cache?
>
>
> No. That wouldn't explain why the Linux browser chokes on any given
> page when the cache is cleared, while the Win2k browser flies when the
> cache is also cleared.
>
> I am convinced there is a DNS lookup problem, where the SBC servers
> don't work smoothly with Linux requests. The technical details of why
> this could be so are beyond my knowledge. I would welcome anyone who
> could explain this, as it would seem incredibly important if such a
> discrepancy is possible, for the community to have collective awareness
> of it and how to affect workarounds.
>
>
> I have optimized my surfing speed experience through other workarounds,
> but none of which resolve the fundamental problem, which persists if I
> leave the browsers in their default configurations, and both OSes wset
> up with the same DNS configurations.
>
> I still wish I could resolve this fundamental problem.
>
>
>
> Good day!


I've been following this issue for some time, and it appears that certain
DNS servers do not catch the first or second request from Linux computers
although it seems they do work properly with MS requests. I have not
confirmed, but I guess there is some small imcompatibility with the DNS
requests. I would bet the servers at fault are running MS and not strictly
following standards.

Chris Carlen wrote:

> Bill Marcum wrote:
>> On Fri, 18 Feb 2005 19:40:08 -0800, Chris Carlen
>> <(E-Mail Removed)> wrote:
>>
>>>Any further suggestions how to improve things would be appreciated.
>>>
>>
>> Increase the size of the browser cache?
>
>
> No. That wouldn't explain why the Linux browser chokes on any given
> page when the cache is cleared, while the Win2k browser flies when the
> cache is also cleared.
>
> I am convinced there is a DNS lookup problem, where the SBC servers
> don't work smoothly with Linux requests. The technical details of why
> this could be so are beyond my knowledge. I would welcome anyone who
> could explain this, as it would seem incredibly important if such a
> discrepancy is possible, for the community to have collective awareness
> of it and how to affect workarounds.
>
>
> I have optimized my surfing speed experience through other workarounds,
> but none of which resolve the fundamental problem, which persists if I
> leave the browsers in their default configurations, and both OSes wset
> up with the same DNS configurations.
>

Does your DNS servers support IPv6? If not check to see if the Mandrake box
is querying with an IPv6 format?

I seemed to have missed the original post. But there are some latency
issues on slow connections because MTU defaults to 1500 byte packets in
linux and more like 256/552 on windows. This results in faster queing of
packets, but more overhead on downloads for windows. Keep in mind that 40
bytes of these packets are tcp/ip header information. Although probably
only an issue on 56K type connections these days.

If your issue is dns, you can verify or at least work around it by
accessing the page by it's IP. Or putting it's <IP> <hostname(s)> in your
/etc/hosts file. If this cures the problem, you can be pretty sure it's a
dns issue. Just bear in mind that many popular sites distribute page
content over several hostnames. Such as login.site images.site
profile.site my.site and others just for one page of content. Which means
you might be doing several DNS requests for one page.

Also the DNS IP's your ISP might have given you, may not be the ones
windows is using. winipcfg or ipconfig /ALL on windows. cat
/etc/resolv.conf in linux. If you're not using the same ones on both
you're comparing apples to oranges. And there's always the IPv4 / IPv6
thing. Not to mention DNS caching with bind/named, dns_masq, squid, ...

HTH,
Shadow_7

>>>>> "r" == ray <(E-Mail Removed)> writes:

r> On Sun, 20 Feb 2005 17:15:22 -0800, Chris Carlen wrote:
>> Bill Marcum wrote:
>>> On Fri, 18 Feb 2005 19:40:08 -0800, Chris Carlen
>>> <(E-Mail Removed)> wrote:
>>>
>>>> Any further suggestions how to improve things would be
>>>> appreciated.
>>>>
>>> Increase the size of the browser cache?
>>
>>
>> No. That wouldn't explain why the Linux browser chokes on any
>> given page when the cache is cleared, while the Win2k browser
>> flies when the cache is also cleared.
>>
>> I am convinced there is a DNS lookup problem, where the SBC
>> servers don't work smoothly with Linux requests. The technical
>> details of why this could be so are beyond my knowledge. I would
>> welcome anyone who could explain this, as it would seem incredibly
>> important if such a discrepancy is possible, for the community to
>> have collective awareness of it and how to affect workarounds.
>>
>>
>> I have optimized my surfing speed experience through other
>> workarounds, but none of which resolve the fundamental problem,
>> which persists if I leave the browsers in their default
>> configurations, and both OSes wset up with the same DNS
>> configurations.
>>
>> I still wish I could resolve this fundamental problem.
>>
>>
>>
>> Good day!


r> I've been following this issue for some time, and it appears that
r> certain DNS servers do not catch the first or second request from
r> Linux computers although it seems they do work properly with MS
r> requests. I have not confirmed, but I guess there is some small
r> imcompatibility with the DNS requests. I would bet the servers at
r> fault are running MS and not strictly following standards.

Would running a caching nameserver help?

Later

ray wrote:
> On Sun, 20 Feb 2005 17:15:22 -0800, Chris Carlen wrote:
>
>
>>Bill Marcum wrote:
>>
>>>On Fri, 18 Feb 2005 19:40:08 -0800, Chris Carlen
>>> <(E-Mail Removed)> wrote:
>>>
>>>
>>>>Any further suggestions how to improve things would be appreciated.
>>>>
>>>
>>>Increase the size of the browser cache?
>>
>>
>>No. That wouldn't explain why the Linux browser chokes on any given
>>page when the cache is cleared, while the Win2k browser flies when the
>>cache is also cleared.
>>
>>I am convinced there is a DNS lookup problem, where the SBC servers
>>don't work smoothly with Linux requests. The technical details of why
>>this could be so are beyond my knowledge. I would welcome anyone who
>>could explain this, as it would seem incredibly important if such a
>>discrepancy is possible, for the community to have collective awareness
>>of it and how to affect workarounds.
>>
>>
>>I have optimized my surfing speed experience through other workarounds,
>>but none of which resolve the fundamental problem, which persists if I
>>leave the browsers in their default configurations, and both OSes wset
>>up with the same DNS configurations.
>>
>>I still wish I could resolve this fundamental problem.
>>
>>
>>
>>Good day!
>
>
>
> I've been following this issue for some time, and it appears that certain
> DNS servers do not catch the first or second request from Linux computers
> although it seems they do work properly with MS requests. I have not
> confirmed, but I guess there is some small imcompatibility with the DNS
> requests. I would bet the servers at fault are running MS and not strictly
> following standards.

Thanks for the input.

This sort of thinking, proposing meaningful hypotheses, is what I am
hoping to elicit here. As I madee clear in my OP, the systems being
compared are configured very much the same, and it appears you have
thoroughly read my OP and so avoided proposing the obvious fixes that
I've already tried.

Thus, observed differences must be caused by some real inconsistencies
between the way the OSes handle network requests, and thus how servers
respond to them.

At this point, I would want to take a theory such as yours, and perform
experiments to try to verify if it is the right explanation. Once the
real problem is found we can try to find a workaround.

I am hoping some networking experts can propose actions to take to
perhaps actually *measure* the DNS resolution times, or perhaps using
tcpdump or other tools, see if there are differences between the way the
two OSes are doing things.


Good day!






--
__________________________________________________ _____________________
Christopher R. Carlen
Principal Laser/Optical Technologist
Sandia National Laboratories CA USA
(E-Mail Removed) -- NOTE: Remove "BOGUS" from email address to reply.

Jerry Smiley wrote:
> Chris Carlen wrote:
>
>
>>Bill Marcum wrote:
>>
>>>On Fri, 18 Feb 2005 19:40:08 -0800, Chris Carlen
>>> <(E-Mail Removed)> wrote:
>>>
>>>
>>>>Any further suggestions how to improve things would be appreciated.
>>>>
>>>
>>>Increase the size of the browser cache?
>>
>>
>>No. That wouldn't explain why the Linux browser chokes on any given
>>page when the cache is cleared, while the Win2k browser flies when the
>>cache is also cleared.
>>
>>I am convinced there is a DNS lookup problem, where the SBC servers
>>don't work smoothly with Linux requests. The technical details of why
>>this could be so are beyond my knowledge. I would welcome anyone who
>>could explain this, as it would seem incredibly important if such a
>>discrepancy is possible, for the community to have collective awareness
>>of it and how to affect workarounds.
>>
>>
>>I have optimized my surfing speed experience through other workarounds,
>>but none of which resolve the fundamental problem, which persists if I
>>leave the browsers in their default configurations, and both OSes wset
>>up with the same DNS configurations.
>>
>
>
> Does your DNS servers support IPv6? If not check to see if the Mandrake box
> is querying with an IPv6 format?


Thanks for the input.

Mandrake box? The problem Linux box is Suse 9.1. As the OP indicates,
the Suse 9.1 box has been adjusted via an edit to /etc/modprobe.conf to
have IPv6 disabled.

But it could still be of value to doublecheck that this edit really
disabled IPv6. And it would also be of value to perform the check that
you propose.

How might I go about verifying that IPv6 is really disabled? And how
can I verify if the DNS queries are being attempted with IPv6 or not?


Good day!







--
__________________________________________________ _____________________
Christopher R. Carlen
Principal Laser/Optical Technologist
Sandia National Laboratories CA USA
(E-Mail Removed) -- NOTE: Remove "BOGUS" from email address to reply.

Shadow_7 wrote:
> I seemed to have missed the original post. But there are some latency
> issues on slow connections because MTU defaults to 1500 byte packets in
> linux and more like 256/552 on windows. This results in faster queing of
> packets, but more overhead on downloads for windows. Keep in mind that 40
> bytes of these packets are tcp/ip header information. Although probably
> only an issue on 56K type connections these days.
>
> If your issue is dns, you can verify or at least work around it by
> accessing the page by it's IP. Or putting it's <IP> <hostname(s)> in your
> /etc/hosts file. If this cures the problem, you can be pretty sure it's a
> dns issue. Just bear in mind that many popular sites distribute page
> content over several hostnames. Such as login.site images.site
> profile.site my.site and others just for one page of content. Which means
> you might be doing several DNS requests for one page.

Yes, that is why it is difficult to make comparisons on complex pages
between the two OSes.

Do you know if it is possible to use some tool to determine all of the
hosts that would be used to render a given web page? For instance, I
can do the command:

> host www.cnn.com

to find the IP of that page, but the other servers called upon when I
actually browse to www.cnn.com are not revealed by that single host lookup.

Is there some command or method by which I can enter a URL, and the tool
or method will spit out a list of ALL servers that need to be contacted
to obtain the data needed to completely render the page?

Thus, with that complete list I could put it into /etc/hosts and be able
to perform a true page rendering with direct IP address accesses (or
more precisely, locally resolved hostnames).


> Also the DNS IP's your ISP might have given you, may not be the ones
> windows is using. winipcfg or ipconfig /ALL on windows. cat
> /etc/resolv.conf in linux. If you're not using the same ones on both
> you're comparing apples to oranges. And there's always the IPv4 / IPv6
> thing. Not to mention DNS caching with bind/named, dns_masq, squid, ...

As I mentioned in the OP,

"The LAN hosts have static IPs on subnet 192.168.1.0 and statically
configured DNS server addresses set to the values shown by the router.
These DNS addresses are always the same: 63.203.35.55
and 206.13.28.12 ."

"The Windows host also has a static IP and manually configured DNS
servers. Also, there is a Win2k VMware machine on each Linux box with
bridged networking and static IP and DNS configurations. "


So I am confident the Windows boxes, both real and virtual are
configured the same. Unless there is any possiblity that Windows can
somehow change its "running" DNS configuration despite being hand
configured with manual DNS server IPs? I doubt even Windows could be
this bonkers. To do so it would seem to require that some special
software be running, such as software that the ISP would provide, but
which *we are not using* because the Win boxes are on a LAN behind a
router, not connected directly to the ISP.

Nonetheless, I will be interested in checking to make sure that the Win
boxes are truly using the DNS config I manually provided. So I will
check this.

Ah, indeed I see no tricky business goes on. When manual DNS servers
are specified, then C:>ipconfig /all in Win2k reveals that those are the
DNS servers actually being used.

So it is confirmed that the Win and Linux machines are using the same DNSes.


Good day!




--
__________________________________________________ _____________________
Christopher R. Carlen
Principal Laser/Optical Technologist
Sandia National Laboratories CA USA
(E-Mail Removed) -- NOTE: Remove "BOGUS" from email address to reply.