Re: [squid-users] help with cachemgr statistics

Thanks very much, Amos.

I will do more math on the statistics i have captured, now having in
mind what you said, just to have a better understanding of the
behaviour of these proxies.

As for the code diving, i'm not sure if i have the necessary skills.

thanks,

Carlos Defoe

On Fri, Mar 29, 2013 at 10:27 AM, Amos Jeffries <squid3_at_treenet.co.nz> wrote:
> On 30/03/2013 12:48 a.m., Carlos Defoe wrote:
>>
>> Hello,
>>
>> I'm investigating cache manager statistics (60min).
>>
>> In one sample, i have:
>> client_http.requests = 22.785904/sec
>> client_http.hits = 6.354478/sec
>> client_http.errors = 0.006388/sec
>> server.http.requests = 16.956939/sec
>> server.http.errors = 0.000000/sec
>> aborted_requests = 0.094992/sec
>>
>> My thoughts:
>>
>> * client_http.requests means all the http request that arrives to squid;
>> * client_http.hits means (approximately) the requests that have been
>> completed without using Internet resource, only getting contents from
>> the cache;
>> * server.http.requests means the requests that are completed making
>> access to webservers and downloading content;
>>
>> So, client_http.hits + server.http.requests =(should be)
>> client_http.requests.
>>
>> But the result is only approximated (23,31 instead of 22,78).
>>
>> 1) How can i write a equation that fits "requests -> completions"?
>> using those errors and aborted requests? i couldn't figure.
>
>
> Unfortunately the recorded metrics are a bit rough and you can't write an
> equation to calculate that with the existing details. Things are not as
> simple as HTTP-in / HTTP-out used to be and the stats counters have not been
> kept up with the feature changes very well.
>
> Squid performs translation from non-HTTP server protocols which arrive as
> client HTTP requests but leave Squid as FTP or Gopher or TCP tunnels.
> Squid also services cachemgr requests and error page embeded content which
> arrive as client HTTP requests but never get near being a HIT *or* MISS.
> Squid also performs re-tries on server errors so one client HTTP request
> MISS can mean multiple server HTTP requests.
>
> * on top of all that, Squid accounting is a bit rough around what is a HIT
> and how REFRESH are accounted for.
>
> So the best metrics to use are actually the client in/out totals and server
> in/out totals for bandwidth. From these you can get a ratio of bandwidth
> reduction or indication of problems.
>
>
>> 2) In another proxy (one very busy, with bad performance),
>> client_http.hits + server.http.requests doesn't go near
>> client_http.requests:
>>
>> client_http.requests = 66.597198/sec
>> client_http.hits = 8.075425/sec
>> client_http.errors = 0.132439/sec
>> server.http.requests = 19.327989/sec
>> server.http.errors = 0.000000/sec
>> aborted_requests = 0.733549/sec
>>
>> In this case, i think most of the requests (66 - (19+8)) are being
>> lost, as this proxy server is in realy bad performance and it's
>> hardware is not capable to serve all the requests. But those lost
>> requests, where have
>> they gone (in statistics)?
>
>
> Good question. Like to go code diving to find out?
>
> Amos
Received on Sun Mar 31 2013 - 01:37:03 MDT