Re: [hylafax-users] what is an acceptable rate of failure ?

[Lee Howard <faxguy@xxxxxxxxxxxxxxxx>]

George H wrote:
> Just what is an acceptable rate of failure when sending/recieving faxes?
>   

It's important to make sure what counts as a failure, and how one then 
calculates a "rate" of its occurrence as compared to what other 
non-failure conditions.

I usually calculate receiving errors as session failures per number of 
pages received (both numbers are easily retrieved via the 'recvstats' 
script).  Some people like to calculate them as session failures per 
total sessions, but I'll explain why I do it as I do...

Normally on a well-tuned system if a session failure occurs with a 
particular sender... the chances of that sender calling back are 
extremely high... and the chances of that subsequent call experiencing 
another similar session failure is also quite high.  So if there is a 
sender with really bad lines (or with VoIP "lines") who wants to send me 
a 10-page fax, and if that fax has a session error occur during every 
other page... then it's going to take 5 or 6 calls to get all of the 
pages through with 4 or 5 session errors.

If I thus calculate receiving error rate as session failures per total 
calls... the error rate is 80% or 83.3% for that caller.

The way that I calculate the receiving error rate (session failures per 
total pages received)... the error rate is 40% or 50% for that caller... 
and this seems a whole lot more accurate to me as far as a meaningful 
representation goes.  The high alternative seems... meaningless, 
really... 80% of the calls fail... but without any indication of how 
long each call was or how much non-error behavior occurred before the 
single error condition... 80% really means very little to me.  40% seems 
a lot more accurate, to me, as far as the representation of our 
communication ability with that sender goes.

Furthermore, it's very easy for a single problematic sender to skew the 
statistics to an incredibly insane degree.  Let me explain...

Assume that you have gotten 100 fax calls with a total of 500 pages in 
them.  Assume that the only errors were one caller who had a hard time 
with a 1-page fax... and they called (and negotiated a "fax" session) 10 
times (this is not an absurd degree of persistence, either).  By 
calculating the error rate per-calls you end up with an error rate of 
10%.  It seems extremely unfair.  Rather, the per-pages rate seems much 
more meaningful, 2%.

Also... realize that HylaFAX doesn't count a receive fax "session" until 
it gets Phase B signaling from the sender.  So if you get telmarketer 
war-dialers pounding on your lines (and all you need is a block of 1000 
DIDs in the USA to see this happen very commonly) who generate 1 
voice-telemarketer call on each DID ... you can see that they will 
greatly skew the error rates by their behavior that has absolutely 
nothing to do with fax or what you're trying to measure.

So that's why I use error rates calculated per-page.  Per-call error 
rates never really mean much to me... they don't seem to measure 
accurately what it is that we're really trying to measure.

So, that said...

On the production systems that I administer I usually see error rates 
(again, per-page) much, much less than 1%.  Usually it's around 0.1%... 
but, depending on that customer's sender pool I've seen it get to what I 
would consider "high" (around 3%, in which cases it's almost always 
because they have a frequent sender who has a bad fax machine or bad 
lines or something).

.....

In sending situations we don't have to worry about voice-calls and other 
similar issues creeping into our statistics ... and although I still do 
calculations on a per-page basis, the per-call information can give you 
a sense for how often people send to wrong numbers or busy numbers or 
such.  Normally I see send error rates around 1%, but this varies widely 
... depending on how trustworthy the numbers are and how frequently we 
send to busy numbers.  Instead of paying any attention to send error 
"rates" I simply look for total job failures (again, if a session fails 
once due to something that is a real problem - something that needs our 
attention - its chances of failing on all attempts is extremely high)...

So in sending conditions if the job failed I ignore any that fail due to 
"busy", "no carrier", or "no answer" conditions.  Everything else I take 
quite seriously.  If I see one such job failure (in an average day of 
sending 1500 pages of fax I pay attention to the session logs to make 
sure that it's not a problem that needs my attention, but I generally 
don't fret too much).  If I see five such job failures in the same 
conditions to different numbers... then I am worried.

Thus... in sending conditions I don't really pay attention to rates... 
but rather whole numbers... I watch the errors themselves.

> I have hylafax+ on 4 lines and using xferfaxlog for the stats I
> calculated we have an average of 86.46% rate of receiving faxes. We're
> talking about maybe 40,000 incomming calls in total.
>   

I'm not sure whether this is calculated per-page or per-call.  If it's 
per-call then this number is really somewhat meaningless to me.  If it's 
per-page then you have problems.

> To me an error is any report from hylafax of an Exx error.
> Success is one where the REASON field in empty.
>
> Also I should mention I live in a country where the phone lines arn't
> exactly perfect. So would there be anything published about it.

If your phone lines are bad then make the telco fix them.  If it's other 
people's from whom you receive faxes frequently, then you should tell 
them about it.

Thanks,

Lee.


____________________ HylaFAX(tm) Users Mailing List _______________________
 To subscribe/unsubscribe, click http://lists.hylafax.org/cgi-bin/lsg2.cgi
On UNIX: mail -s unsubscribe hylafax-users-request@xxxxxxxxxxx < /dev/null
 *To learn about commercial HylaFAX(tm) support, mail sales@xxxxxxxxx*