From: Thayne Harbaugh <tharbaugh@lnxi.com>
To: netdev@oss.sgi.com
Cc: ganesh.venkatesan@intel.com
Subject: [PATCH] abysmal e1000 performance (DITR)
Date: Tue, 01 Jun 2004 17:05:12 -0600 [thread overview]
Message-ID: <1086131111.20113.22.camel@tubarao> (raw)
The introduction of Dynamic Interrupt Throttling Rate (DITR) in the 4.x
-> 5.x e1000 driver change can cause serious performance problems. The
ITR can be reduced when the system is *not* under load which results in
gratuitous latencies for network traffic. In other words - why is the
interrupt load reduced when the system isn't under load?
This is a patch to include system load when calculating the DITR. It
only allows the diff/goc ratio to factor into the DITR when the 1 minute
load average is above .50. It appears to work quite well and prevents
throttling when the load is below .50 and allows throttling when the
load is in excess of .50.
There are a few concerns. The patch requires kernel/timer.c to
EXPORT_SYMBOL(avenrun) - otherwise the symbol isn't available to the
driver when built as a module. This symbol "clutter" may be undesirable
to some. Another concern is that NAPI accomplishes a similar interrupt
load reduction and is likely a better solution than ITR.
diff -ur linux-2.4.21-99/drivers/net/e1000/e1000_main.c linux-2.4.21-99-e1000/drivers/net/e1000/e1000_main.c
--- linux-2.4.21-99/drivers/net/e1000/e1000_main.c 2003-09-24 05:47:33.000000000 -0700
+++ linux-2.4.21-99-e1000/drivers/net/e1000/e1000_main.c 2004-05-27 23:38:02.000000000 -0700
@@ -27,6 +27,8 @@
*******************************************************************************/
#include "e1000.h"
+/* for avenrun[] */
+#include <linux/sched.h>
/* Change Log
*
@@ -1429,14 +1431,34 @@
/* Dynamic mode for Interrupt Throttle Rate (ITR) */
if(adapter->hw.mac_type >= e1000_82540 && adapter->itr == 1) {
+ /* fixed fractional part of .50 */
+#define FIXED_F50 (FIXED_1 >> 1)
+ /* This maps the range of .50-.99 -> 0-100 */
+#define FIXED_1_MAPPED (FIXED_1 - FIXED_F50)
+ unsigned long laf; /* load average fractional part */
+ uint32_t goc; /* good octet count */
+ uint32_t dif; /* dif between tx and rx goc */
+ uint32_t itr; /* inturrept throttle rate */
+ /* laf range is mapped:
+ * .00-.50 -> .00
+ * .50-.99 -> .00-.99
+ * 1.00+ -> 1.00 */
+ if (avenrun[0] + FIXED_1/200 > FIXED_1)
+ laf = FIXED_1_MAPPED;
+ else if (avenrun[0] + FIXED_1/200 < FIXED_F50)
+ laf = 0;
+ else
+ laf = ((avenrun[0] + FIXED_1/2000) & (FIXED_1 - 1)) - FIXED_1_MAPPED;
/* Symmetric Tx/Rx gets a reduced ITR=2000; Total
* asymmetrical Tx or Rx gets ITR=8000; everyone
* else is between 2000-8000. */
- uint32_t goc = (adapter->gotcl + adapter->gorcl) / 10000;
- uint32_t dif = (adapter->gotcl > adapter->gorcl ?
- adapter->gotcl - adapter->gorcl :
- adapter->gorcl - adapter->gotcl) / 10000;
- uint32_t itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
+ goc = (adapter->gotcl + adapter->gorcl) / 10000;
+ dif = (adapter->gotcl > adapter->gorcl ?
+ adapter->gotcl - adapter->gorcl :
+ adapter->gorcl - adapter->gotcl) / 10000;
+ itr = goc > 0
+ ? 8000 + 6000*laf/FIXED_1_MAPPED*dif/goc - 6000*laf/FIXED_1_MAPPED
+ : 8000;
E1000_WRITE_REG(&adapter->hw, ITR, 1000000000 / (itr * 256));
}
diff -ur linux-2.4.21-99/kernel/Makefile linux-2.4.21-99-e1000/kernel/Makefile
--- linux-2.4.21-99/kernel/Makefile 2003-09-24 05:47:27.000000000 -0700
+++ linux-2.4.21-99-e1000/kernel/Makefile 2004-05-26 23:45:47.000000000 -0700
@@ -11,7 +11,7 @@
export-objs = signal.o sys.o kmod.o context.o ksyms.o pm.o exec_domain.o \
printk.o cpufreq.o rcupdate.o syscall_ksyms.o fork.o hook.o \
- rashooks.o module.o
+ rashooks.o module.o timer.o
obj-y = sched.o dma.o fork.o exec_domain.o panic.o printk.o \
module.o exit.o itimer.o info.o time.o softirq.o resource.o \
diff -ur linux-2.4.21-99/kernel/timer.c linux-2.4.21-99-e1000/kernel/timer.c
--- linux-2.4.21-99/kernel/timer.c 2003-09-24 05:47:27.000000000 -0700
+++ linux-2.4.21-99-e1000/kernel/timer.c 2004-05-27 00:54:29.000000000 -0700
@@ -686,6 +686,7 @@
* all seem to differ on different machines.
*/
unsigned long avenrun[3];
+EXPORT_SYMBOL(avenrun);
static inline void calc_load(unsigned long ticks)
{
--
Thayne Harbaugh
Linux Networx
next reply other threads:[~2004-06-01 23:05 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2004-06-01 23:05 Thayne Harbaugh [this message]
2004-06-04 14:20 ` [PATCH] abysmal e1000 performance (DITR) Thayne Harbaugh
-- strict thread matches above, loose matches on Subject: below --
2004-06-04 15:44 Venkatesan, Ganesh
2004-08-26 17:55 ` jamal
2004-08-26 20:04 ` Thayne Harbaugh
2004-08-26 20:26 ` Jeff Garzik
2004-08-26 21:28 ` Thayne Harbaugh
2004-08-28 22:58 ` Jeff Garzik
2004-08-27 21:49 Ronciak, John
2004-08-27 22:05 ` Thayne Harbaugh
2004-08-27 22:41 Ronciak, John
2004-08-30 17:06 ` Thayne Harbaugh
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1086131111.20113.22.camel@tubarao \
--to=tharbaugh@lnxi.com \
--cc=ganesh.venkatesan@intel.com \
--cc=netdev@oss.sgi.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).