OS X 10.4.2, storing objects roughly over 32K ...

[Jason Edgecombe]

Hi James,

Is this Python?

Have you tried using a different memcached client like Perl?
I seem to recall someone saying that the python client wasn't great or 
something like that.

Jason

James Robinson wrote:

> To update myself, I have gotten things rolling based on advice found:
>
>     http://lists.danga.com/pipermail/memcached/2005-March/001292.html
>
> With set_cork() defined as:
>
> void set_cork(conn *c, int val) {
>    if (c->is_corked == val) return;
>    c->is_corked = val;
>
>    val = ~val;
>    setsockopt(c->sfd, IPPROTO_TCP, TCP_NODELAY, &val, sizeof(val));
> }
>
> stores of payload between 14000 and 195846 bytes all happen on OS X  
> without delay. Apparently OS X's TCP_NOPUSH is just plain busted, as  
> seems to be the net wisdom garnered from google.
>
> ---
> import memcache
> import time
> import socket
>
> c = memcache.Client(['127.0.0.1:11211'])
>
> for i in (32641, 32639, 16000, 14000, 18000, 32641, 32644,55000,  
> 32641 * 2, 32641 * 4, 32641 * 6):
>     val = 's' * i
>     t1 = time.time()
>     c.set(i,val)
>     t2= time.time()
>     print "%d store took %s seconds" % (i, t2  - t1)
> ---
>
> 32641 store took 0.00501680374146 seconds
> 32639 store took 0.00244212150574 seconds
> 16000 store took 0.00187420845032 seconds
> 14000 store took 0.00256204605103 seconds
> 18000 store took 0.00198698043823 seconds
> 32641 store took 0.00225687026978 seconds
> 32644 store took 0.00342202186584 seconds
> 55000 store took 0.00297403335571 seconds
> 65282 store took 0.0147829055786 seconds
> 130564 store took 0.00387501716614 seconds
> 195846 store took 0.025102853775 seconds
>
>
>
> ----
> James Robinson
> Socialserve.com
>
>
>