CentOS - Oct 2017 - Weird bandwith behaviour (download throughput) on CentOS based gateway

On Thu, Oct 05, 2017 at 02:57:18PM +1300, Clint Dilks (clintd at
scms.waikato.ac.nz) wrote:
> On Thu, Oct 5, 2017 at 2:41 PM, Jobst Schmalenbach <jobst at
barrett.com.au>
> wrote:
> [snip]
> Hi,
> 
> Are you sure  that your issue  isn't related to the mirror that your
> systems are selecting ?  If they are using different mirrors I would try
> using the fastestmirror plugin to make the gateway select the same mirror
> as you other host.
Darn!!!!

I should have included in the initial email that I actually ran EXTRA tests from
a {local} CentOS mirror using wget after I figured there was some differences in
the "yum update" times ...

The Gateway:

  [root at GATEWAY /tmp] #>wget
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
  --2017-10-05 14:59:55-- 
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
  Resolving mirror.internode.on.net... 150.101.135.3
  Connecting to mirror.internode.on.net|150.101.135.3|:80... connected.
  HTTP request sent, awaiting response... 200 OK
  Length: 3972005888 (3.7G) [application/octet-stream]
  Saving to: ???CentOS-6.9-x86_64-bin-DVD1.iso???
  0% [                                                                          
] 4,454,198    680K/s

One of the hosts behind it:

  [root at piquet /tmp] #>wget
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
  --2017-10-05 15:01:32-- 
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
  Resolving mirror.internode.on.net... 150.101.135.3
  Connecting to mirror.internode.on.net|150.101.135.3|:80... connected.
  HTTP request sent, awaiting response... 200 OK
  Length: 3972005888 (3.7G) [application/octet-stream]
  Saving to: `CentOS-6.9-x86_64-bin-DVD1.iso'
  0% [                                                                          
] 13,616,730  2.4M/s


Jobst


-- 
You have junk mail.

  | |0| |   Jobst Schmalenbach, General Manager
  | | |0|   Barrett & Sales Essentials
  |0|0|0|   +61 3 9533 0000, POBox 277, Caulfield South, 3162, Australia

Jobst Schmalenbach <jobst at barrett.com.au> writes:
> On Thu, Oct 05, 2017 at 02:57:18PM +1300, Clint Dilks (clintd at
scms.waikato.ac.nz) wrote:
>> On Thu, Oct 5, 2017 at 2:41 PM, Jobst Schmalenbach <jobst at
barrett.com.au>
>> wrote:
>> [snip]
>> Hi,
>> 
>> Are you sure  that your issue  isn't related to the mirror that
your
>> systems are selecting ?  If they are using different mirrors I would
try
>> using the fastestmirror plugin to make the gateway select the same
mirror
>> as you other host.
>
> Darn!!!!
>
> I should have included in the initial email that I actually ran EXTRA
> tests from a {local} CentOS mirror using wget after I figured there
> was some differences in the "yum update" times ...
>
> The Gateway:
>
>   [root at GATEWAY /tmp] #>wget
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
>   --2017-10-05 14:59:55-- 
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
>   Resolving mirror.internode.on.net... 150.101.135.3
>   Connecting to mirror.internode.on.net|150.101.135.3|:80... connected.
>   HTTP request sent, awaiting response... 200 OK
>   Length: 3972005888 (3.7G) [application/octet-stream]
>   Saving to: ???????CentOS-6.9-x86_64-bin-DVD1.iso??????
>   0% [                                                                     
] 4,454,198    680K/s
>
> One of the hosts behind it:
>
>   [root at piquet /tmp] #>wget
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
>   --2017-10-05 15:01:32-- 
http://mirror.internode.on.net/pub/centos/6.9/isos/x86_64/CentOS-6.9-x86_64-bin-DVD1.iso
>   Resolving mirror.internode.on.net... 150.101.135.3
>   Connecting to mirror.internode.on.net|150.101.135.3|:80... connected.
>   HTTP request sent, awaiting response... 200 OK
>   Length: 3972005888 (3.7G) [application/octet-stream]
>   Saving to: `CentOS-6.9-x86_64-bin-DVD1.iso'
>   0% [                                                                     
] 13,616,730  2.4M/s
Is there a dependency on which machine you test first?  Perhaps the file
has been stored in some cache along the way and for the second test, it
can be delivered from the cache instead of from the source, which might
yield higher speeds.


-- 
"Didn't work" is an error.

On Mon, Oct 09, 2017 at 03:05:51PM +0200, hw (hw at adminart.net)
wrote:
> Jobst Schmalenbach <jobst at barrett.com.au> writes:
> 
> > On Thu, Oct 05, 2017 at 02:57:18PM +1300, Clint Dilks (clintd at
scms.waikato.ac.nz) wrote:
> >> On Thu, Oct 5, 2017 at 2:41 PM, Jobst Schmalenbach <jobst at
barrett.com.au>
> 
> Is there a dependency on which machine you test first?  Perhaps the file
> has been stored in some cache along the way and for the second test, it
> can be delivered from the cache instead of from the source, which might
> yield higher speeds.
Very good question, answer is no for the following reasons:

 - it happens for all downloads - yum, wget etc
 - I have looked at the interfaces using ngrep, all traffic goes straight out
through the closest (as in hops) interface
 - As you raised this I have disabled caching on the command line using wget,
still happens
 - As you raised this I have checked whether there are any (environment) options
set, none
   I, too, use the same bash scripts on all machines I have

I though about the interfaces, but can't be. The last two interfaces are on
the problem machine, but when downloading on the LAN I get a throughput of
~28mbs, only when downloading on the gateway I only get <10mbs.

So still baffled.

Jobst




-- 
Computing power increases as the square of the cost. If you want to do it twice
as cheaply, you have to do it four times as fast.

  | |0| |   Jobst Schmalenbach, General Manager
  | | |0|   Barrett & Sales Essentials
  |0|0|0|   +61 3 9533 0000, POBox 277, Caulfield South, 3162, Australia