Re: [squid-users] Re: is there any thing wrong from cache manager logs ?!!

On 11/11/2013 01:16 PM, Amos Jeffries wrote:
> On 11/11/2013 10:08 p.m., Dr.x wrote:
<SNIP>
Amos just a tiny question at the end..

>> hi amos , thanks alot .
>>
>> im trying to understand the calculation equation why it is as the formula u
>> posted above ??
>> not understanding ,
>> ""bytes from client ""
>> from client to wt ?? from client to squid ?
>>
>> ==================================================
>> can i calculate the speed savings??
>>
>> i mean that i want to calculate the bw over speed not over bytes
>>
>> also i want to see this saving every 60 sec
>> is that possible ?
>
> Sure. Poll the SNMP every 60sec and calculate.
>
OK true but not exactly.
In this context I want to ask something:
will squid ever change any object time related headers?
such as Last-Modified or Expires etc??
let say I am getting lots of 304 requests suddenly after changing couple
squid refresh_patterns will it mean that squid saved me some bandwidth
that was not marked by a HIT??

In a manner of speed squid is a cache which can benefit since it's smart.
Lets a assume that dumb Goethe's "The Sorcerer's Apprentice" router does
everything he was told to do..
All the requests from the clients will be served in a case of router
since it does everything the client forces him\it to do.
In the case of Squid and Squid only (as a service) it will consume the
bandwidth smartly!
which means that not every request that the client will want to fetch as
fresh will be fetched as from from origin server but only these that has
the characteristics for that.
It's not perfect but it has a very smart logic compared to a dumb router..
So yes you can count in and out bytes etc in a manner of 700MB..
The only thing is that a Gb interface can go two ways (IN\OUT) 1Gbps (1
Giga Bits Per Second) while 700MB is about 5600Mbps which I have not
seen ethernet in this speed..(yes I noted that you have mentiend Mbps
and not MB..)

So it's not an equation that you just put three numbers in it and thats it.
you need to know that squid is the only service that is running on this
machine if you calculate it based on the interface.
If there is any other service on squid machine looking at the interface
will lead to unreliable statistics.
You need to know all squid dns cost in bytes, all squid http requests
cost in bytes and then using the whole squid http "server side" bytes
towards the client and the access.logs output and only then... determine
whether putting squid gave any benefit to the client bandwidth and or
lowering the costs of bandwidth.

If and when you set your goals to make clients surfing speed faster
using squid gives you the basic option to start allowing it.
While using no squid at all leaves you with no option at all.
Note that there are cases which squid machine can lead to some slowdowns
while allowing some other fastenings of surging speed.

Eventually if you will only look at SNMP,
(AMOS: just a quick and unsure thing, what SNMP OIDS are SMP supported?)
What you will see is bytes and this is not how you look at HTTP traffic
at all.

Until now I have not seen a log analyzer for squid 3.2+ logs which
contains couple new TCP_X flags which turns the tables while calculating
BYTES..
Like for example: every request that the client forcibly asked for a
fresh object means that the client understood or felt that there is a
need to not relay on either the browser cache or the intermediate cache,
logic and decision.
So in a case you have total requests of 300k and 90% of them was forced
by the clients to be fetched fresh you can try to see what and why this
is happening and maybe understand that most of your users do not even
use any content that caching fits to at all.
Note that the above test is only to showing that HTTP cache is not an
equation you can say about "in" vs "out" and you need to also analyze
the access.log to make sure that there is some traffic patterns from the
clients that a central cache will benefit you for them.

I hope that the above was informative for you in a way and leaves the
question:
What SNMP counters can be used with SMP?

Thanks,
Eliezer

>
>> ===========================================
>> i use mrtg monitors and usually i have basic snmp on interface
>> i look at traffic speed between "in" , "out" , speed"in"-speed"out"=speed
>> saving ( i know its not accurate)
>> my question is does my calcultion bw saving speed containg a very error
>> value ?
>
> You are using the word "speed" here in strange ways.
>
> speed is a measurement of something over time.
>
> How many seconds can you have in 1 second? dont make much sense to have
> more than 1 or less than 1. So how can you save some seconds of the 60
> sec period? ...
>
> The thing which is changing is the bytes. You get more or less total
> bytes in each direction every second (or every 60 seconds).
>
> Proxy does not gain in speed, but it does gain/lose in some bytes each
> time period.
>
> So you take the calculation I gave for "bandwidth saving/loss" every 60
> seconds. Also subtract the total savings value you had 60 seconds ago,
> and the result is the savings in just these latest 60 seconds.
>
> I assume that is what you are meaning. Since the other translations do
> not make sense at all.
>
>> assume i have 700Mbps pumped to squid.
>> =======================================================
>> regards
>>
>
>
> Amos
>
Received on Mon Nov 11 2013 - 14:11:12 MST