To: ptb@inv.it.uc3m.es
cc: "Alvaro Herrera" <alvherre@commandprompt.com>,
	"Richard Huxton" <dev@archonet.com>, pgsql-performance@postgresql.org
Subject: Re: general PG network slowness (possible cure) (repost) 
In-reply-to: <200705251545.l4PFjf612758@inv.it.uc3m.es> 
References: <200705251545.l4PFjf612758@inv.it.uc3m.es>
Comments: In-reply-to "Peter T. Breuer" <ptb@inv.it.uc3m.es>
	message dated "Fri, 25 May 2007 17:45:41 +0200"
Date: Fri, 25 May 2007 12:16:21 -0400
Message-ID: <11030.1180109781@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"Peter T. Breuer" <ptb@inv.it.uc3m.es> writes:
> "Also sprach Tom Lane:"
>> Except that in the situation you're describing, there's only a hundred
>> or two bytes of response to each query, which means that only one send()
>> will occur anyway.  (The flush call comes only when we are done
>> responding to the current client query.)

> It may still be useful. The kernel won't necessarily send data as you
> push it down to the network protocols and driver. The driver may decide
> to wait for more data to accumulate,

No, because we set TCP_NODELAY.  Once we've flushed a message to the
kernel, we don't want the kernel sitting on it --- any delay there adds
directly to the elapsed query time.  At least this is the case for the
final response to a query.  I'm not too clear on whether this means we
need to be careful about intermediate message boundaries when there's a
lot of data being sent.

			regards, tom lane