Re: [gentoo-user] I/net server throttling

[Mick]

On Friday 22 Jun 2012 11:29:00 William Kenworthy wrote:
> On Fri, 2012-06-22 at 05:10 -0400, Philip Webb wrote:
> > 120622 Mick wrote:
> > > On Friday 22 Jun 2012 05:41:14 Philip Webb wrote:
> > >> In recent weeks, I've seen many instances of downloads of files
> > >> coming in small segments, which seem to be the same for each server,
> > >> but vary between different servers.  Some Gentoo mirrors suffer this
> > >> way -- not all -- & some general news sites, eg when delivering
> > >> videos, which run for a few secs, then freeze, then resume etc.
> > >> Once I observed a server doing this for several hours (a large file),
> > >> then suddenly loosening up & delivering the rest of the file in one
> > >> go.
> > >
> > > Yes, when I use the wireless network at work.
> > > It could be the ISP or it could be other users taking up bandwidth.
> > > Really early in the morning performance is generally higher
> > > than during work hours.  this could be due to contention
> > > on the local network or the ISP's pipe.
> > > Do you get such problems off peak hours?
> >
> > I think you're describing a different phenomenon, traffic jams (smile).
> > That sometimes happens here & may persist for  1 - 2 weeks ,
> > but it feels just like driving on the highway when  1  lane is closed.
> > That probably is some piece of the Internet or ISP under repair or test.
> >
> > What I'm refering to is getting downloaded files in slices,
> > eg using Wget, a piece of the file downloads for  22 sec , then stops;
> > Wget tries again & another  22 sec  piece comes down the pipe, then
> > stops. This can go on for hours with a big file
> > & doesn't seem related to the local time of day.
> > My guess is that the server has been programmed to stop after  22 sec
> > in an effort to share access among many clients,
> > but it is irritating & also suggests the server needs faster hardware.
> >
> > Have others noticed this -- it seems to be a recent innovation --
> > & is it a known ploy of server managers ?
>
> Are you using traffic shaping like a "police filter", or is there
> shaping somewhere in the path? - most of the linux shaping methods work
> on a burst principle that produces a given throughput by gating the
> traffic for an average throughput.  I have seen the effect you mention
> with block transfer protocols (ftp) and wget when trying to use its
> built in bandwidth regulation which works similarly to when traffic
> shaping is in effect.  There seems to be an interaction between the two
> instances of regulating the traffic to create an effect like you have
> seen.  This also affects other traffic trying to use the link at the
> same time as the available bandwidth gets very "choppy".
>
> I have less experience with other than linux based shaping methods such
> as Cisco QoS methods but have not seen the same effect there.

Now that I understood what Philip was describing I have not seen anything like 
that (yet).  When there is throttling at the server it just shaves off 
anything above a certain transmission rate.  So a bar chart that would 
otherwise show a variable bar height as the client-server try to negotiate 
maximum throughout up to the ADSL max capacity, is now cropped at a lower than 
max throughput.  There is no starting and stopping.

If there is contention on the line then there will be some throughput 
variability, but not really stopping completely and then restarting.  Unless 
perhaps the contention is so abysmal or the server/ISP are under some DDoS 
attack?

-- 
Regards,
Mick