"Why "using the same 5000 handles" ? Sure, if all those 5000 are already
completed then sure you can reuse them but I would guess than some of those
transfers aren't completed in a second so you need to create and use more
handles while the previous ones are finishing up."

Allow me to correct myself. By 5000 handles I meant, 5000 TCP connections /
5000 open sockets no more no less. This is a fixed requirement.

So basically in order to achieve this,
1. I should fix the number of TCP connections to 5000 by setting it using
CURLMOPT_MAXCONNECTS.
2. If there are no handles available, I can create a easy handle on the fly
and use that to send the request.
3. So I can in fact in one second can have 5000 TCP concurrent connections
sending 15000 requests using 15000 handles.
4. This also means that for a single connection, before we have received
200 OK for a request, we would have sent another request which is what I am
looking for.

Correct me if I am wrong.

Regards,

On Thu, Sep 18, 2014 at 3:58 PM, Daniel Stenberg <daniel_at_haxx.se> wrote:

> On Thu, 18 Sep 2014, nikhil ap wrote:
>
> Set up 5000 concurrent connections and vary the request rate from 5000 to
>> 50000 requests per second.
>>
>
> libcurl doesn't really expose "connections" in the API, but I guess we can
> ignore that for now.
>
> Assuming I have setup 5000 easy handles and added it to multi interface,
>> in
>> the first second, 5000 GET requests will be sent.
>>
>
> Yeps.
>
> In the next second, using the same 5000 handles, I would like to send
>> 10000 GET requests.
>>
>
> Why "using the same 5000 handles" ? Sure, if all those 5000 are already
> completed then sure you can reuse them but I would guess than some of those
> transfers aren't completed in a second so you need to create and use more
> handles while the previous ones are finishing up.
>
> This means 2 requests should be pipelined and sent in a single handle.
>>
>
> Not exactly. The second request could possibly be pipelined on an existing
> connection if you have that enabled, but the second request needs its own
> easy handle. There's a one-to-one mapping between transfers and easy
> handles. The second GET in a pipelined connection is still a separate
> transfer and thus it is made with a separate easy handle. The multi handle
> is what keeps track of all connections and all possible pipelines.
>
> In the third second, I would like to send 15000 which would have piplined
>> 3
>> requests and so on.
>>
>
> Again, you then need 15000 unused easy handles to take care of those 15000
> transfers. Some of them may then end up being done pipelined, some of them
> may not.
>
>
> --
>
> / daniel.haxx.se
> -------------------------------------------------------------------
> List admin: http://cool.haxx.se/list/listinfo/curl-library
> Etiquette: http://curl.haxx.se/mail/etiquette.html
>

-- 
niks

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2014-09-18