Re: Query random paths?

[Jan W. <J.W...@cs...>]

On Monday 27 July 2009 07:38:12 am Robert Iakobashvili wrote:
> Hi Jan,
>
> On Fri, Jul 24, 2009 at 4:25 PM, Jan Wielemaker 
<J.W...@cs...>wrote:
> > Hi,
> >
> > Just trying my first steps into curl-loader :-)  I looked through the FAQ
> > and config files, but I couldn't find an answer on how to query a *lot*
> > of random paths using a pool of -lets say- 20 clients.
> >
> > I see we have FORM_RECORDS_RANDOM, which seems to do the trick for
> > form-data. I am trying to debug server crashes, where I know it is
> > related to concurrency, but otherwise I have little clue on whether or
> > not it is related to the actual query paths.
> >
> > Just to bash the server, I collected a list of 6,000 paths that have
> > been queried on the server recently and then I use the config below, but
> > curl-loader only seems to use the first 20 paths of the file :-(
> >
> > Did I overlook something trivial?
> >
> >        Cheers --- Jan
> >
> > ================================================================
> > BATCH_NAME= bulk
> > CLIENTS_NUM_MAX=20
> > CLIENTS_NUM_START=20
> > CLIENTS_RAMPUP_INC=0
> > INTERFACE=eth0
> > NETMASK=16
> > IP_ADDR_MIN= 192.168.1.1
> > IP_ADDR_MAX= 192.168.1.200
> > IP_SHARED_NUM=200
> > CYCLES_NUM= -1
> > URLS_NUM= 1
> >
> > URL_TEMPLATE=http://localhost:3040%s
> > URL_TOKEN_FILE=paths
> > URL_SHORT_NAME="local-server"
> > REQUEST_TYPE=GET
> > TIMER_URL_COMPLETION = 5000
> > TIMER_AFTER_URL_SLEEP = 0
> > ================================================================
>
> It is a rather recent feature developed by Gary and it will be great if
> Gary will comment.

I've seen that the config allows for getting random records from form-data
files.  Would be nice to have the same for token-files.

> Still, you are keeping 20 clients, which are testing 20 paths.
> Could it be that 6000 clients will do the job? Give it a try.

Probably that will work, but I don't want a lot of concurrent clients.

> Just keep in mind, that by default you need for 35K of memory per each
> client
> and ramp-up about 100 not to start all them at start-up.

I must admit that I still don't really get the meaning of the CIENTS_*
settings.  Is the number of clients expressed by this formula (t is
time in seconds).

	max(CLIENTS_NUM_MAX, CLIENTS_NUM_START+t*CLIENTS_RAMPUP_INC)

	--- Jan