From PKadziolka at acppharma.pl  Sun Sep  7 12:52:00 2008
From: PKadziolka at acppharma.pl (Piotr Kadziolka)
Date: Sun, 7 Sep 2008 14:52:00 +0200
Subject: how to compute measured bwidth
Message-ID: <OF0045C944.5301B64E-ONC12574BD.0043FFCE-C12574BD.00472E45@acppharma.pl>

Hi,

I've obtained below results from report generator. I attached load trace 
also. My question is how to compute measured bandwidth because when I plot 
rep.rate multiplied by rep.size.mean I can't obtain trace similar to this 
at pict. 

According to results offered load in volume is higher than measured but 
when I plot req.rate multiplied by rep.size.mean (which theoretically 
shuld be offered load) together with measured load (rep.rate x 
rep.size.mean) I got the same trace. Please, tell me how to compute 
measured load in volume context.

Load            Count           Volume
                (xact/sec)      (Mbits/sec)
offered         150.34          10.53
measured        150.33          3.35



Thanks, Peter
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.web-polygraph.org/pipermail/users/attachments/20080907/460afbb8/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: load.trace.scope=sides=client__phases=framp,fill,fexit,inc1,top1,dec1,idle,inc2,top2,dec2
Type: application/octet-stream
Size: 6677 bytes
Desc: not available
URL: <http://lists.web-polygraph.org/pipermail/users/attachments/20080907/460afbb8/attachment.obj>

From PKadziolka at acppharma.pl  Sun Sep  7 23:46:45 2008
From: PKadziolka at acppharma.pl (Piotr Kadziolka)
Date: Mon, 8 Sep 2008 01:46:45 +0200
Subject: Odp: Re: how to compute measured bwidth
In-Reply-To: <1220819819.13755.1192.camel@pail>
Message-ID: <OF0CABD163.F72FF844-ONC12574BD.00826D33-C12574BD.00831FAC@acppharma.pl>

Hi Alex,

I had deep look at binary log and I know why bwidth dropped. Near at the 
end of fill phase I saw xact error and from that point to the end of test 
I observed a lot lot errors. I count almost 70 percent contribution of 
errors in totals transactions. It was one type of error: (c19) unsupported 
HTTP status code.

766.81| Xaction.cc:74: error: 1/2 (c19) unsupported HTTP status code
1108188903.706303# obj: 
http://w1033.h1130o1101s1010.bench.tst/w0baba0e0.3a2d01ab:00000430/t02/_00000513.html 
flags: rpt,basic,GET, size: 0/-1 xact: 0baba0e0.3a2d01ab:0068626e
HTTP/1.1 502 Proxy Error ( W magazynie brak miejsca dla wykonania tego 
polecenia.  There was no free space in the storage to execute that 
command)
Via:1.1 SR-ISA
Pragma: no-cache
Cache-Control: no-cache
Content-Type: text/html
Content-Length: 3795

One of two disk had crashed. That's all mystery.

Thanks for Your help Alex,
Peter



Alex Rousskov <rousskov at measurement-factory.com> 
2008-09-07 23:45

To
Piotr Kadziolka <PKadziolka at acppharma.pl>
cc
users at web-polygraph.org
Subject
Re: how to compute measured bwidth






On Sun, 2008-09-07 at 14:52 +0200, Piotr Kadziolka wrote:

> I've obtained below results from report generator. I attached load
> trace also. My question is how to compute measured bandwidth because
> when I plot rep.rate multiplied by rep.size.mean I can't obtain trace
> similar to this at pict. 
> 
> According to results offered load in volume is higher than measured
> but when I plot req.rate multiplied by rep.size.mean (which
> theoretically shuld be offered load) together with measured load
> (rep.rate x rep.size.mean) I got the same trace. Please, tell me how
> to compute measured load in volume context. 
> 
> Load                Count                Volume 
>                 (xact/sec)        (Mbits/sec) 
> offered         150.34                10.53 
> measured        150.33                3.35 
> 

Interesting. Can you send me (off-list) the binary logs from this test?
I will investigate why the bandwidth dropped while request rate remained
the same.

Thank you,

Alex.



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.web-polygraph.org/pipermail/users/attachments/20080908/0210a172/attachment.html>

From OskarS at flashnetworks.com  Wed Sep 17 19:57:04 2008
From: OskarS at flashnetworks.com (Oskar Svadkovskey)
Date: Wed, 17 Sep 2008 22:57:04 +0300
Subject: ERROR: "PolyApp.cc:720: (s22) Invalid argument"
Message-ID: <95962D8FA988994E8268F6B2889C812329A2FD5CBC@ILmail01.flashnetworks.com>

I've just installed polygraph on FreeBSD using instructions in http://www.measurement-factory.com/docs/PolyMix-4/.

Preliminary tets tests seem to be running fine. I have no idea if I should ignore this error and, if not, how I can diagnose the problem.

_________________________________________

PolySrv# ./poly.sh pg-workloads/simple.pg
/usr/local/polygraph/bin/polysrv: warning: no run phases were specified; generating and using default phase
PolyApp.cc:720: (s22) Invalid argument
000.01| Command: /usr/local/polygraph/bin/polysrv --cfg_dirs pg-workloads/include --verb_lvl 10 --config pg-workloads/simple.pg
000.01| Configuration:
        version:            2.8.1
        host_type:          i386-unknown-freebsd6.3
        verb_lvl:           10
        dump:               err,
        dump_size:          1.000KB
        notify:             <none>
        doorman_listen_at:  <none>
        doorman_send_to:    <none>
        label:              [none]
        fd_limit:           10749
        config:             pg-workloads/simple.pg
        cfg_dirs:           pg-workloads/include
        console:            -
        log:                [none]
        log_buf_size:       -1Bytes
        store_working_set:  [none]
        load_working_set:   [none]
        sample_log:         [none]
        sample_log_buf_size:-1Bytes
        accept_foreign_msgs:off
        stats_cycle:        5.00sec
        file_scan:          poll
        priority_sched:     5
        fake_hosts:
        delete_old_addrs:   yes
        idle_tout:          5.00min
        local_rng_seed:     1
        global_rng_seed:    1
        unique_world:       on
        ign_urls:           off
000.01| Server content distributions:
        Server S101:
                content   planned%    likely%     error% mean_sz_bytes
        some-content                   100.00     100.00       0.00   13326.41
        expected average server-side cachability: 80.00%
        expected average server-side object size: 13326.41Bytes

000.01| Phases:
     phase  pop_beg  pop_end load_beg load_end  rec_beg  rec_end smsg_beg smsg_end     goal      flags
dflt              1.00     1.00     1.00     1.00     1.00     1.00     1.00     1.00 <none>

000.01| StatsSamples:
        static stats samples:   0
        dynamic stats samples:  0

000.01| FDs: 11095 out of 11095 FDs can be used; safeguard limit: 10749
000.01| resource usage:
        CPU Usage: 22msec sys + 701msec user = 723msec
        Maximum Resident Size: 14.488MB
        Page faults with physical i/o: 0

000.01| group-id: 12701342.646e43a7:00000004 pid: 17319
000.01| current time: 1221680528.725671 or Wed, 17 Sep 2008 19:42:08 GMT
000.01| fyi: PGL configuration stored (370bytes)
000.01| fyi: no bench selected with use(); will not attempt to create agent addresses
000.01| created 1 agents total
000.01| Server S101 [1 / 12701342.646e43a7:00000006] at 127.0.0.1:9090 HTTP/1.1
000.01| fyi: current state (1) stored
000.01| starting 1 HTTP agents...
000.09| i-dflt      0   0.00     -1  -1.00   0    1
^Cgot shutdown signal (2)
000.11| noticed shutdown signal (2)
000.11| resource usage:
        CPU Usage: 22msec sys + 704msec user = 726msec
        Maximum Resident Size: 14.488MB
        Page faults with physical i/o: 0

000.11| fyi: current state (2) stored
000.11| server 127.0.0.1:9090 is closing listen socket 3 after 0 xactions
000.11| got 0 xactions and 0 errors
000.11| shutdown reason: got shutdown signal

PolySrv#




DISCLAIMER:

This e-mail message and its attachments contain information that may be privileged or confidential and is the property of Flash networks Ltd. This message is only intended for the person to whom it is addressed, therefore if you are not the intended recipient, please note that you are not authorized to read, print, retain, copy disseminate, distribute, or use this message or any part thereof. If you received this message in error, please notify the sender immediately and delete all copies of this message. Any views expressed in this message are those of the individual sender, except where the sender specifically states them to be the views of Flash Networks Ltd.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.web-polygraph.org/pipermail/users/attachments/20080917/300787ac/attachment.html>

From rousskov at measurement-factory.com  Thu Sep 18 02:13:03 2008
From: rousskov at measurement-factory.com (Alex Rousskov)
Date: Wed, 17 Sep 2008 20:13:03 -0600
Subject: ERROR: "PolyApp.cc:720: (s22) Invalid argument"
In-Reply-To: <95962D8FA988994E8268F6B2889C812329A2FD5CBC@ILmail01.flashnetworks.com>
References: <95962D8FA988994E8268F6B2889C812329A2FD5CBC@ILmail01.flashnetworks.com>
Message-ID: <1221703983.6395.44.camel@pail>

On Wed, 2008-09-17 at 22:57 +0300, Oskar Svadkovskey wrote:
> I?ve just installed polygraph on FreeBSD using instructions in
> http://www.measurement-factory.com/docs/PolyMix-4/. 
> 
> Preliminary tets tests seem to be running fine. I have no idea if I
> should ignore this error and, if not, how I can diagnose the problem.

> PolyApp.cc:720: (s22) Invalid argument
> 
> 000.01| Command: /usr/local/polygraph/bin/polysrv --cfg_dirs
> pg-workloads/include --verb_lvl 10 --config pg-workloads/simple.pg

Ignore it. It is a minor problem: Polygraph failed to find a way to
setup a handler for the "out of RAM" condition. That handler does not
affect Polygraph behavior during a test but may help save more results
if the process runs out of RAM due to overload or misconfiguration. The
next release will have a more robust handler setting code.

HTH,

Alex.




From OskarS at flashnetworks.com  Thu Sep 18 10:57:48 2008
From: OskarS at flashnetworks.com (Oskar Svadkovskey)
Date: Thu, 18 Sep 2008 13:57:48 +0300
Subject: unknown option "NMBCLUSTERS"
Message-ID: <95962D8FA988994E8268F6B2889C812329A2FD5ED6@ILmail01.flashnetworks.com>

I had a problem compiling the kernel with the NMBCLUSTERS option. I eventually compiled the kernel without it. I understand that this option is relevant only for older kernels. Is this correct? (I am using FreeBSD 6.3-RELEASE)

options MAXFILES=16384
options HZ=1000
options NMBCLUSTERS=32678
POLYGRAPH: 293 lines, 10823 characters.

PolySrv# config POLYGRAPH
POLYGRAPH: unknown option "NMBCLUSTERS"
PolySrv#






DISCLAIMER:

This e-mail message and its attachments contain information that may be privileged or confidential and is the property of Flash networks Ltd. This message is only intended for the person to whom it is addressed, therefore if you are not the intended recipient, please note that you are not authorized to read, print, retain, copy disseminate, distribute, or use this message or any part thereof. If you received this message in error, please notify the sender immediately and delete all copies of this message. Any views expressed in this message are those of the individual sender, except where the sender specifically states them to be the views of Flash Networks Ltd.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.web-polygraph.org/pipermail/users/attachments/20080918/5e7c73c9/attachment.html>

From wessels at measurement-factory.com  Thu Sep 18 16:33:52 2008
From: wessels at measurement-factory.com (Duane Wessels)
Date: Thu, 18 Sep 2008 10:33:52 -0600 (MDT)
Subject: unknown option "NMBCLUSTERS"
In-Reply-To: <95962D8FA988994E8268F6B2889C812329A2FD5ED6@ILmail01.flashnetworks.com>
References: <95962D8FA988994E8268F6B2889C812329A2FD5ED6@ILmail01.flashnetworks.com>
Message-ID: <20080918103233.L71618@measurement-factory.com>




On Thu, 18 Sep 2008, Oskar Svadkovskey wrote:

> I had a problem compiling the kernel with the NMBCLUSTERS option. I eventually compiled the kernel without it. I understand that this option is relevant only for older kernels. Is this correct? (I am using FreeBSD 6.3-RELEASE)

Yes, For newer FreeBSD versions it is no longer configurable in the kernel config file.

However you can set it in /boot/loader.conf like this if you need to:

    kern.ipc.nmbclusters=32768

Duane W.


From OskarS at flashnetworks.com  Fri Sep 19 21:45:25 2008
From: OskarS at flashnetworks.com (Oskar Svadkovskey)
Date: Sat, 20 Sep 2008 00:45:25 +0300
Subject: unknown option "NMBCLUSTERS"
In-Reply-To: <20080918103233.L71618@measurement-factory.com>
References: <95962D8FA988994E8268F6B2889C812329A2FD5ED6@ILmail01.flashnetworks.com>
	<20080918103233.L71618@measurement-factory.com>
Message-ID: <95962D8FA988994E8268F6B2889C812329A2FD6191@ILmail01.flashnetworks.com>


On Thu, 18 Sep 2008, Oskar Svadkovskey wrote:

> I had a problem compiling the kernel with the NMBCLUSTERS option. I eventually compiled the kernel without it. I understand that this option is relevant only for older kernels. Is this correct? (I am using FreeBSD 6.3-RELEASE)

Yes, For newer FreeBSD versions it is no longer configurable in the kernel config file.

However you can set it in /boot/loader.conf like this if you need to:

    kern.ipc.nmbclusters=32768

Duane W.

************************

Just to make sure that I get you right. 6.3-RELEASE is using one of the newer kernels?

Sorry for my ignorance. We've installed FreeBSD only because of web-polygraph.  I am not familiar with this system at all. And we are on this, I see another version of web-polygraph in ports. It's not the stable and not the new. It's something in between. I just can't get to the system right now to tell you exactly what I see. So I am a bit confused as to whether I should stick to the stable version that's on the site or I should use one in ports.



DISCLAIMER:

This e-mail message and its attachments contain information that may be privileged or confidential and is the property of Flash networks Ltd. This message is only intended for the person to whom it is addressed, therefore if you are not the intended recipient, please note that you are not authorized to read, print, retain, copy disseminate, distribute, or use this message or any part thereof. If you received this message in error, please notify the sender immediately and delete all copies of this message. Any views expressed in this message are those of the individual sender, except where the sender specifically states them to be the views of Flash Networks Ltd.




From wessels at measurement-factory.com  Sat Sep 20 02:49:29 2008
From: wessels at measurement-factory.com (Duane Wessels)
Date: Fri, 19 Sep 2008 20:49:29 -0600 (MDT)
Subject: unknown option "NMBCLUSTERS"
In-Reply-To: <95962D8FA988994E8268F6B2889C812329A2FD6191@ILmail01.flashnetworks.com>
References: <95962D8FA988994E8268F6B2889C812329A2FD5ED6@ILmail01.flashnetworks.com>
	<20080918103233.L71618@measurement-factory.com>
	<95962D8FA988994E8268F6B2889C812329A2FD6191@ILmail01.flashnetworks.com>
Message-ID: <20080919204622.G71618@measurement-factory.com>




On Sat, 20 Sep 2008, Oskar Svadkovskey wrote:

> Just to make sure that I get you right. 6.3-RELEASE is using one of the newer kernels?

Yes, thats correct.


Add the nmbclusters setting to /boot/loader.conf (from previous email)
and you'll be fine.

> So I am a bit confused as to whether I should stick to the stable
> version that's on the site or I should use one in ports.

It probably depends on what Polygraph features you want to use.  Polygraph
doesn't have any external dependencies (I think) so installing from source
is not much harder than installing from ports.

Duane W.


From OskarS at flashnetworks.com  Sun Sep 28 13:25:27 2008
From: OskarS at flashnetworks.com (Oskar Svadkovskey)
Date: Sun, 28 Sep 2008 16:25:27 +0300
Subject: Addressing scheme
Message-ID: <95962D8FA988994E8268F6B2889C81232A87288CD6@ILmail01.flashnetworks.com>

Hello List

I have a question regarding the addressing for polymix4. It's said that the bench id should be between 100 and 199. It should be the second octet in address spaces and the third in host IPs. Due to limitations of our current network configuration I simply can't use something between 100 and 99 in these octets of the address spaces. What I am missing here??? I  see the customized pg file  runs ok and creates load without benchmark ID.

oskar




DISCLAIMER:

This e-mail message and its attachments contain information that may be privileged or confidential and is the property of Flash networks Ltd. This message is only intended for the person to whom it is addressed, therefore if you are not the intended recipient, please note that you are not authorized to read, print, retain, copy disseminate, distribute, or use this message or any part thereof. If you received this message in error, please notify the sender immediately and delete all copies of this message. Any views expressed in this message are those of the individual sender, except where the sender specifically states them to be the views of Flash Networks Ltd.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.web-polygraph.org/pipermail/users/attachments/20080928/a7ad37a0/attachment.html>

From rousskov at measurement-factory.com  Tue Sep 30 20:19:39 2008
From: rousskov at measurement-factory.com (Alex Rousskov)
Date: Tue, 30 Sep 2008 14:19:39 -0600
Subject: Addressing scheme
In-Reply-To: <95962D8FA988994E8268F6B2889C81232A87288CD6@ILmail01.flashnetworks.com>
References: <95962D8FA988994E8268F6B2889C81232A87288CD6@ILmail01.flashnetworks.com>
Message-ID: <1222805979.21272.133.camel@pail>

On Sun, 2008-09-28 at 16:25 +0300, Oskar Svadkovskey wrote:

> I have a question regarding the addressing for polymix4. It?s said
> that the bench id should be between 100 and 199. It should be the
> second octet in address spaces and the third in host IPs. Due to
> limitations of our current network configuration I simply can?t use
> something between 100 and 99 in these octets of the address spaces.
> What I am missing here??? I  see the customized pg file  runs ok and
> creates load without benchmark ID.

The strict rules were designed for cache-offs and other product
comparisons where it was very important that all participants know what
to expect and can expect similar testing conditions.

You should customize the workloads to fit your environment as necessary.

Good luck,

Alex.