Re: [squid-users] equal bandwidth

From: Alexander Dudko <ibaba@dont-contact.us>
Date: Mon, 13 May 2002 21:27:22 +0200

On Mon, 13 May 2002 09:52:13
 Squid Support (Henrik Nordstrom) wrote:
>On Monday 13 May 2002 03:22, Alexander Dudko wrote:
>
>> delay_pools 1
>>
>> delay_class 1 2
>> delay_access 1 allow allowed_hosts
>> delay_access 1 deny all
>> delay_parametrs 1 16000/16000 400/400
>
>This will give each user ca 3kbit, and not more than 128kbit in total
>of all users.
>
>> Did I make it correctly?
>
>You probably want to leave the per-user limit somewhat higher with
>some seconds bucket size to allow for burst, and the total limit
>somewhat smaller to avoid saturating the link.
>
>Note: You quite likely also need to tune down the TCP window size, at
>least on Linux. 8KB is probably a reasonable value for what you are
>doing here.
>
>> As I know, squid will divide all bandwidth between all clients.
>> But... what it takes place when some users are turned off, and
>> don't use their part of traffic? Will another users's bandwidth
>> increase more than 400, or no? And by what way?
>
>The least available amount of both pool limits apply.
>
>> And finally... I have squid-2.4STABLE6 on RedHat Linux.
>> Maybe the new version of squid have new features to work with
>> bandwidth?
>
>2.4.STABLE6 is fine. There is no additional delay pool features in
>later Squids (yet).
>
 
Thank you.

But what is about TCP window size? Why it is likely? Could you explain or give some links?

Any I don't understand: "The least available amount of both pool limits apply".
If I have following configuration, for example:
delay_pools 1
delay_class 1 2
delay_access 1 allow allowed_hosts
delay_access 1 deny all
delay_parametrs 1 16000/16000 400/400

And there are 40 users in our network. But now 20 of them is absent.
In this case, how many byte per second does each present user have, in reality?
Can squid automatically increase the bandwidth of each present user in order to fill all 16000 bytes of incoming traffic?

        Best regards,

                        Alexander Dudko.
Received on Mon May 13 2002 - 13:27:35 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:08:06 MST