A couple of questions about report

Dmitry Kurochkin dmitry.kurochkin at measurement-factory.com
Fri Dec 9 21:06:06 UTC 2011


On Fri, 9 Dec 2011 13:37:40 -0700, Chengyu Fan <chengy.fan at gmail.com> wrote:
> Thanks for reply so soon.
> 
> On Fri, Dec 9, 2011 at 7:11 AM, Dmitry Kurochkin <
> dmitry.kurochkin at measurement-factory.com> wrote:
> 
> > Hi Chengyu.
> >
> > On Thu, 8 Dec 2011 15:17:10 -0700, Chengyu Fan <chengy.fan at gmail.com>
> > wrote:
> > > Hi~ all,
> > > Sorry I have some naive questions again.
> > > I modify the simple download test in my experiment. And I generate the
> > > report using polygraph-reporter.
> > > I am a little confused about some parts in the report. At the end of this
> > > test, the polygraph client cannot connect to the server side, and exits.
> > > The generated report is attached. The questions are as follows:
> > >
> > > 1. The load trace graph shows the offered rate and measured rate, but it
> > > only based on the client-side.
> >
> > If you mean the load trace figure on the "traffic rates, counts, and
> > volumes" page, then yes.
> >
> > > So the client cannot distinguish the
> > > measured rate comes from the proxy or the polygraph server, right?
> >
> > It can.  Otherwise Polygraph would not be able to calculate Hit ratio.
> >
> > > Is there
> > > a graph we can directly check how many rate comes from proxy and how many
> > > comes from the server?
> > >
> >
> > If proxy is used, all replies the client receives come from the proxy,
> > clients do not talk directly to the origin servers in this case.  Also I
> > am not sure I understand what "how many rate" means.
> >
> 
> For "how many rates", I mean right now we know the measured rate from the
> client side. But I don't know among all these measured rate, how many
> measured rate comes from  server, and how many comes from proxy.
> Is there a way to do that?
> 

To collect detailed hits and misses stats, Polygraph needs the proxy to
forward custom headers (X-Xact at least) in requests and replies.

> Also, I noticed that in the load trace, the measured rate is not always
> there. Sometimes I can only see the offered information. Why is this?
> 

Apparently, in your test the proxy removes extension headers from
forwarded messages.  That is why Polygraph can not collect detailed hits
and misses stats.

> 
> >
> > There are "traffic rates, counts, and volumes" page has client-side
> > reply traffic stream table at the end.  It contains contribution and
> > rates stats for various reply types, including "misses" and "hits and
> > misses".  That may be what you are looking for.  You can click on
> > individual reply type to see detailed stats for it.
> >
> > Also the Details link on the index page goes to the index of all
> > collected stats for all phases, both client and server sides.
> >
> 
> Yeah, I checked that. But in the traffic stream table, I found that the
> misses is always 100%. But I do use a proxy, does this mean that the proxy
> does not cache anything? And I checked log file, the server side rate is
> much small than client sends. I think it is the proxy replies requests
> instead of server. So why the misses is 100%?
> 

I tried to explain why this happens above.  In your test the proxy does
caching, you can see the aggregated hit ratio in the first table on the
"hit ratios" page.  But detailed hits and misses stats can not be
collected.

> 
> >
> > > 2. In hit ratios part, I find the measured number in the first table is
> > > more than 79% (I set the recurrence is 80%). But in the client side hit
> > > ratios, the measured number is 0. Why is that?
> > >
> >
> > Polygraph relies on custom HTTP headers to calculate detailed hit
> > stats.  If for some reason Polygraph can not get these headers
> > (e.g. proxy does not forward it), detailed hit ratio stats can not be
> > calculated.
> >
> > Polygraph still can measure the average hit ratio based on client and
> > server stats (the first table on the "hit ratios" page).
> >
> > > 3. Also, I find the server side has lots of errors like "foreign host
> > name"
> > > and "missing target info in the request", does that mean the proxy does
> > not
> > > work well? Or this is the polygraph deliberately to do so?
> > >
> >
> > "Missing target info in the request" means that proxy did not forward
> > X-Target request header.
> >
> > "Foreign host name" means that Polygraph server received request for an
> > unknown host (either in the request URI or the Host header).  Since you
> > are using a single server, that means that proxy changed the host in
> > some way.
> >
> > These issues are likely related to the missing client side stats
> > problem.
> >
> 
> Do you mean, "foreign host name" and "missing target info in the request"
> may be the reasons that there are such a high misses and no enough measured
> information in my report?
> 

"Foreign host name" errors seem like a separate issue.  I do not know
why the proxy changes host in forwarded requests but it is weird.

As for the "missing target info in the request" errors, they are not the
reason, they are the consequence.  The root cause is that the proxy
strips extension headers from the forwarded messages.  If you fix that,
the errors would be gone and you would get detailed hits and misses
stats.

Regards,
  Dmitry

> 
> >
> > > 4. I am also do not know what's the meaning of fill. Could you please
> > > explain it?
> > >
> >
> > Fill is a cachable miss, i.e. a reply that may increase (fill) the proxy
> > cache.
> >
> > Regards,
> >  Dmitry
> >
> > > Thanks.
> > >
> > > --
> > > Regards,
> > > Chengyu Fan
> > > _______________________________________________
> > > Users mailing list
> > > Users at web-polygraph.org
> > > http://www.web-polygraph.org/mailman/listinfo/users
> >
> 
> 
> 
> -- 
> Regards,
> Chengyu Fan



More information about the Users mailing list