On Wed, 30 Aug 2000, Ittai Gilat wrote:
> We are trying to run the webaxe-1.pg workload in a reverse proxy
> configuration. There are problems we bumped into:
>
> - Polygraph client robots send direct web requests (with no http://
> prefix in the URL) only when --proxy is not used. Therefore, client
> webaxe-1.pg files should set
> srv_ips = <reverse-proxy-ip-address>
> which is not a polygraph server. Server webaxe-1.pg files should set
> srv_ips = <polygraph-server-ip-address>
> causing the workload files on the client and server to be different.
Ittai,
Since servers do not really care about Robot configurations, I
think you can have identical PGL files on both sides by adding a "fake"
Server bound to reverse-proxy-ip-address and using that server in Robot
configurations. The fake server PGL configuration must be the same as
all other servers. Also, you will need to tell all servers to use the
same random sequences using the "--unique_world yes --rng_seed 123457"
options for polysrvs. The RNG seed needs to be changed between
experiments unless you clean your cache.
We used the approach above for L4 tests and it worked. You will
get some oid errors though.
> Does this fact have any effect on the test? I tried it, and it worked.
It should work. However, you will receive errors that you describe
below.
Is it possible to use --proxy option instead? While not ideal, it may be
a cleaner solution. Ideally, we need to add a "virtual server" support
to Polygraph to eliminate workarounds and errors.
> - Polygraph robots detect the fact that they do not receive the response
> directly from a polygraph server. I suppose they are able to do so using
> one of the special polygraph X-...: headers. When this happens, I get
> a "foriegn name" error on the client. Does this have any effect on the
> test?
This error will probably go away if you use the fake server approach
above. It will probably have an effect if you are using embedded objects
(they will not be recognized).
It looks like it is better not to use embedded objects if you are not
using the --proxy option.
> - We are currently running a test with one polygraph server. The hit ratio
> we see is very low - 30-40% instead of 0.75 * 0.95 (cacheable *
> hit-ratio).
> I figure that webaxe-1.pg has the same problems mix3 had after moving
> to 2.5.4. Am I right?
I am not sure what problems you are referring to (there were many prior
to 2.5.4). WebAxe-1 does not have IMS requests and reloads, so most
early 2.5 bugs are not applicable. At any rate, 2.5.4 fixed all known
hit ratio bugs.
> - How do we simulate a configuration with more than 1 hosted server?
> To be able to do this, I need the ability to tell a robot that two
> web server names are on the same ip address, right? or is there another
> way...
I think you have two options:
- use --proxy option
- use "fake" Server that is identical to all other servers
except for the IP address.
IIRC, HTTP/1.1 requires Web servers to accept http://host/ URLs. Is
there a reason you do not want to use the --proxy option?
Please note that I have not run many WebAxe tests myself, so I may be
missing something obvious. As far as I know, a few folks are using
WebAxe successfully though (possibly with some tweaks I am not aware
about). Please keep us posted on your progress.
Thanks,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:15 MDT