Re: [squid-users] Corrupt downloads via squid cache

From: Alan J. Flavell <flavell@dont-contact.us>
Date: Mon, 13 May 2002 11:03:17 +0100 (BST)

On Sun, 12 May 2002, Henrik Nordstrom wrote:

> And unfortunately the two reports we have received on this both were
> for others proxies, not their own.. Without the support from the
> staff running the faulty installations there is only so much we can
> do,

Obviously, that is well understood, which is why I sent you a courtesy
email to explain the situation while we tried to progress that; and
indeed you will see some progress on that shortly.

> To verify the case: Have you verified that the problem is indeed at
> the campus/ISP cache, and not caused locally?

Definitely, yes. I'm sorry if that was not clear from the previous
discussion.

Let me put this in detail.

We have two networks (let's call them B and C).

Network B is mandated to use the campus cache for all external access.
Network C is permitted to make external access directly.

We thus have the following scenarios

1. Network B or C -> campus cache -> remote server

2. Network B or C -> local cache -> campus cache -> remote server

3. Network C -> local cache -> remote server.

and of course 4. Network C -> direct to remote server.

In situation 3, no corruption of this type has been seen in the
wild nor have we succeeded in provoking it. Nor in situation 4 of
course.

(The only problems we saw with our own cache involved partial
byteranges - specifically, PDFs, but we're not talking about that
here.)

Most of the samples of corrupted files were gathered using scenario
1, i.e withOUT the local cache being involved.

We had tried at least two different clients (wget and lynx) and
at least two different host computers.

Some samples have also been gathered using scenario 2, but our
conclusion is that the corruption was effectively the same as in
scenario 1, i.e had been caused by the campus cache rather than by the
local cache.

Of course, the local cache is lightly loaded, whereas the campus cache
is heavily loaded - at times so heavily loaded as to cause user
complaints.

> I.e. do you see the
> same kind of corruption when using the campus/ISP cache directly,
> bypassing any local proxies..

That's a definite "yes".

We'll be getting further details to you shortly.

cheers
Received on Mon May 13 2002 - 04:03:20 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:08:05 MST