RE: is IPC really faster than TCPIP?
- From: "Cary Millsap" <cary.millsap@xxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Thu, 2 Sep 2004 11:09:34 -0500
I hear there's a train in France that can go almost 200mph. It may =
improve
the performance of my life a little bit before I'm dead, because someday
I'll probably ride it once. The existence of that fast train will make =
less
than a 1% difference in the performance of my life.
There's undoubtedly someone in France who used to ride in a much slower
train to work everyday, and his two-hour commute now consumes only =
twelve
minutes at 200mph. The performance of that 2-hour period improved by =
10x.
"How much faster" something gets is strictly a function of response =
time.
You can define response time as a measurement of anything you want, but =
when
you communicate with the business, you better measure response time in =
terms
of wall time that a user (a person) experiences.
So, it's common for an IPC connection to process a single dbcall =
round-trip
about 50x faster than a TCP/IP connection on a LAN (from 0.01s to =
0.0002s).
If 92% of a user's response time is spent making round-trips across =
TCP/IP
to begin with, then the 50x improvement in round-trip times will result =
in a
10x improvement in performance. (Try it; run the numbers.)
If only 5% of a user's response time is spent making round-trips across
TCP/IP to begin with, then you can brag about your so-called "50x
improvement" all you want, but the MOST that the user is EVER going to =
see
is a 5% performance improvement in response time.
So, in some cases "no real difference" is right. In some cases "10x" is
right. In other cases, "10,000x" is right. It depends. What it depends =
UPON
is what proportion of your time you spend using the improved resource. =
It's
Amdahl's Law. And this is why focusing on RESPONSE TIME is so important.
Cary Millsap
Hotsos Enterprises, Ltd.
http://www.hotsos.com
* Nullius in verba *
Upcoming events:
- Performance Diagnosis 101: 9/14 San Francisco, 10/5 Charlotte, 10/26
Toronto
- SQL Optimization 101: 9/20 Hartford, 10/18 New Orleans
- Hotsos Symposium 2005: March 6-10 Dallas
- Visit www.hotsos.com for schedule details...
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx =
[mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Juan Carlos Reyes Pacheco
Sent: Wednesday, September 01, 2004 4:24 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: is IPC really faster than TCPIP?
Hi list, Sorry for starting two thread about the same, but I found=20
Tom Kyte says "well, tuning TCP vs IPC would give you marginal =
improvements.
http://asktom.oracle
com/pls/ask/f?p=3D4950:8:::::F4950_P8_DISPLAYID:6136118136754
=20
but Kenny Smith in Use IPC for local connections, a process can be up to =
10x
faster
he says " I've seen a SQL job that runs in 10 minutes using TCP on a =
local
machine run as fast as one minute using an IPC connection"
But http://searchoracle.techtarget.com/originalContent/0,289142
sid41_gci940662,00.html
=20
So I think the question is not is why ipc is not faster, else is IPC =
really
faster.
Some experiences please.=20
=20
Juan Carlos Reyes Pacheco
OCP
----------------------------------------------------------------
To unsubscribe send an email to: oracle-l-request@xxxxxxxxxxxxx
and put 'unsubscribe' in the subject line.
----------------------------------------------------------------
---
To unsubscribe - mailto:oracle-l-request@xxxxxxxxxxxxx&subject=unsubscribe
To read recent messages - //freelists.org/archives/oracle-l/09-2004
Other related posts:
