Re: [tlug] Packet loss: How much is normal, and who should I talk to

[Jim Breen <jimbreen@example.com>]

On 21 February 2014 14:21, Edmund Edgar <lists@example.com> wrote:

> So I'm wondering:
> - Where I see a fair bit of packet loss before I get to the US, is that
> something exceptionally bad that shows there's already a problem at my end /
> with USEN, or is it par for the course?

Hmmm. Remember that the main reason for packet loss is congestion. Queues
overflow in routers and packets get dumped. Not a big issue - it's the way the
Internet is designed to work. The TCP modules at the ends of the connections
throttle back until it clears, and all that most users experience is a period of
slowness. Testers like ping plug on, which is why you see the losses.

Generally when you see a step function in the losses, you've just reached the
place where the loss is happening, so when they jumped before you got to
Rackspace, you'd encountered a problem along the way.

> - When it gets to the Rackspace systems and we start getting 99.4% packet
> loss listed, does that pretty much conclusively show there's a problem at
> their end, or could that show up as the result of failures closer to me?
> Alternatively, is it something that could happen with a normally-functioning
> system for whatever reason?

No, it'll be a problem in their outfit. Probably someone has picked
that moment to
shift a few Gigabytes through one of their switches, and a lot of packets got
trampled on.

If it's happening often scream and yell at them. Threaten to go elsewhere, blab
the story to the press, etc.

I use a Rackspace host and I do a lot of text work on it, so I'm
shipping a lot of
packets back and forth with single-byte payloads. I often see 10-20
second stalls,
and I've suspected it was congestion in their network, but I've never got around
to looking at it further. Maybe I should scream and yell too.

Cheers

Jim

-- 
Jim Breen
Adjunct Snr Research Fellow, Japanese Studies Centre, Monash University