[9fans] iostats ?

[9fans] iostats ?

[steve.simon]

Hi,

Is iostats broken or am I missing somthing?

	iostats ls /bin /bin/* > /dev/null

reports zero average time for all 9p operations, and

	read      85629 bytes, 8.362207e+07 Kb/sec
	write     6401 bytes, 6250977 Kb/sec
	protocol  93942 bytes, 9.174023e+07 Kb/sec

and some impressive throughput rates...

I am running on a 233Mhz P1 so perhaps my clock tick rate
is confused?

-Steve

Re: [9fans] iostats ?

[rog]

> 	iostats ls /bin /bin/* > /dev/null
> reports zero average time for all 9p operations, and
[...]
> and some impressive throughput rates...

it's been broken, quite recently it seems.
in /sys/src/cmd/iostats/statsrv.c:/^update change

<	ulong t2;
<
<	t2 = msec();
<	t -= t2;

to

>	t = msec() - t;

then the numbers become non-zero, at least.

the e+07 numbers are due to the dodgy "avoid zero devide" hack
in iostats.c:

	brpsec = (float)stats->totread / (((float)rpc->time/1000.0)+.000001);
	bwpsec = (float)stats->totwrite / (((float)rpc->time/1000.0)+.000001);
	bppsec = (float)stats->nproto / ((ttime/1000.0)+.000001);

maybe it'd be less misleading just to do:

	setfcr(getfcr() & ~FPZDIV);

before the calculations, and remove the .000001 fudge factor: then if
there were no ops the throughput comes out as +Inf.

Re: [9fans] iostats ?

[steve.simon]

[-- Attachment #1: Type: text/plain, Size: 58 bytes --]

Thanks for that,

All working as expected now.

-Steve

[-- Attachment #2: Type: message/rfc822, Size: 1356 bytes --]

From: 9fans@cse.psu.edu
To: 9fans@cse.psu.edu
Subject: Re: [9fans] iostats ?
Date: Thu, 24 Jul 2003 15:25:14 +0100
Message-ID: <398344982@snellwilcox.com>

> 	iostats ls /bin /bin/* > /dev/null 
> reports zero average time for all 9p operations, and
[...]
> and some impressive throughput rates...

it's been broken, quite recently it seems.
in /sys/src/cmd/iostats/statsrv.c:/^update change 

<	ulong t2;
<
<	t2 = msec();
<	t -= t2;

to

>	t = msec() - t;

then the numbers become non-zero, at least.

the e+07 numbers are due to the dodgy "avoid zero devide" hack
in iostats.c:

	brpsec = (float)stats->totread / (((float)rpc->time/1000.0)+.000001);
	bwpsec = (float)stats->totwrite / (((float)rpc->time/1000.0)+.000001);
	bppsec = (float)stats->nproto / ((ttime/1000.0)+.000001);

maybe it'd be less misleading just to do:

	setfcr(getfcr() & ~FPZDIV);

before the calculations, and remove the .000001 fudge factor: then if
there were no ops the throughput comes out as +Inf.