-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 13/02/2015 8:05 a.m., Yuri Voinov wrote: > I want to say: > > A system with low consumpted memory will starts processes more > quickly against system with low free memory and overloaded cheduler > and more sleep processes in most cases. > > So, IMHO better solution is possibility to shutdown idle children > after configured timeout and starts them up when required. > > Also, OS cache also still exists..... and can hold rewriter ELF > near CPU.... did you remember? ;)
It is a tradeoff. The forking method Squid currently uses causes the child process to get as much virtual memory allocated as the Squid worker process was using at the time, including a full copy of the cache_mem contents. In a low memory system its better to start a bunch of the helpers via startup=N before the parent allocates too much memory and keep them running than it is to start later via idle=N when traffic is going through. Now that you know your peak usage is around 60 helpers worth of traffic you may want to check out what the memory profile is like with startup=50. Amos -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.22 (MingW32) iQEcBAEBAgAGBQJU3TqnAAoJELJo5wb/XPRjyxoIAN5zCpFoaIt/SEpor0WA1U/6 rdG4dSKO5rYBwglHqAKxNLh13nB2MHxMfasHjcje8Z1W59+E/EIYuy6mrCyGTiBm Ckr1wg6JrUcnRQRx7OJBO+9pd6ONtg6wJx4jGarE3kVbHDoccWKhmk60m/LQiPSJ UuXBdEiDPSJW7FV2wEWekp+nZhsIwIgm2xgF6IyVUqYBZ8MsNIeIwJx5z2oJVwaB dloQgExgaCgD23L6j7KzwWfCZsc1JA8pefdjXHYlUTpekUxHB6KZzAUFrSPLokov D8LauV2nXhRl+eVWXAz9uV8bbN6kPbN6GbKdD6eFCtIwwkpy4oOSwsj/UqPUUe0= =52ZF -----END PGP SIGNATURE----- _______________________________________________ squid-users mailing list squid-users@lists.squid-cache.org http://lists.squid-cache.org/listinfo/squid-users
