Re: [squid-users] Delay pools bucket refill

From: Johannes Buchner <buchner.johannes_at_gmx.at>
Date: Fri, 26 Dec 2008 18:14:40 +0100

Hi all!

I tested delay pools to answer my question using this configuration:

1> acl night time 1:00-7:00
2> acl testtime time 17:46-17:48
3> delay_pools 2
4> delay_class 1 1 # full bandwith
5> delay_class 2 2 # limited per user
6> delay_parameters 1 -1/-1
7> delay_parameters 2 -1/-1 1000/1000000
8> delay_access 1 allow night testtime all
9> delay_access 2 allow !night !testtime all
A> delay_initial_bucket_level 100

- Download a file of 51 MB at 17:44
- first 1 MB is full speed (from pool 2, line 7). full speed is 400KB/s
here.
- then the speed is 1000B/s (from pool 2, line 7).
- at 17:46 the download continues to go with 1000B/s. The pool is not
switched.
- I am restarting the transfer: full speed for the whole
transfer. If it takes longer than the time-acl is limited to, the full
speed is still given.

Conclusion: The pools are used connectionwise, and do not switch over
when the time-acl becomes active. In other words, the delay_pools are
chosen by looking at the acls only at the beginning of the transfer. I
assume this is a wise design decision, as continueous checks would
decrease performance.

It means however, that one can start a huge transfer in the 'night' and
block the traffic during the day. The effect can be minimized by using
request_body_max_size.

To my original question:
If you drained your bucket at the end of the acl time, how full is it
at the beginning of the acl time the next time. Is it full, is it
delay_initial_bucket_level, is it stored from the last fill state, does
it refill in the time the acl is inactive?

My test showed that it in fact refills during the time the acl is
inactive. If the acl is inactive 3 minutes (if testtime is 3 minutes
long), the bucket should be filled with 1000*3*60 Bytes, and my tests
confirmed this.

I assume the calculation behind it is invoked on a new connection and
just looks at the time passed since the last transfer and calculates
the new fill state from the timespan. Calculating in the time-acls
would be too complex, as the acls can be configured in quite complex
ways.

Best regards,
Johannes

On Wed, 24 Dec 2008 15:47:12 +1300
Amos Jeffries <squid3_at_treenet.co.nz> wrote:

> Johannes Buchner wrote:
> > Hi!
> >
> > I have a question about delay_pools: If I make a time-based acl
> > with a delay-pool, does it refill in the time the acl is inactive
> > or is the amount "stopped" and continued when the acl starts again?
>
> Pools refill at the constant rate unless the are full or
> reconfigured. Client usage is not taken into consideration on the
> filling, only on the emptying.
>
> >
> > Like, if I have a pool acl going from 9:00 till 20:00 with a size
> > of 3GB and a rate of 1200 B/s, and a client runs low on the bucket
> > at 20:00. What will he be able to download at 9:00 the next day?
> >
> > Also, if I would define one bucket for 9:00 till 20:00 and another
> > one for 20:00 till 9:00 of different sizes and rates, would they
> > share their amount? Probably not I guess.
>
> Correct. No. They are different pools.
>
> Amos
> --
> Please be using
> Current Stable Squid 2.7.STABLE5 or 3.0.STABLE11
> Current Beta Squid 3.1.0.3

-- 
Emails können geändert, gefälscht und eingesehen werden. Signiere oder
verschüssele deine Mails mit GPG.
http://web.student.tuwien.ac.at/~e0625457/pgp.html

Received on Fri Dec 26 2008 - 17:15:02 MST

This archive was generated by hypermail 2.2.0 : Fri Dec 26 2008 - 12:00:01 MST