linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [RFC] Poor Network Performance with e1000 on 2.6.14.3
@ 2006-02-01  5:17 Ashutosh Naik
       [not found] ` <43E04712.6080108@candelatech.com>
  0 siblings, 1 reply; 2+ messages in thread
From: Ashutosh Naik @ 2006-02-01  5:17 UTC (permalink / raw)
  To: linux-kernel, netdev, linux-net, Andrew Morton, jgarzik
  Cc: john.ronciak, ganesh.venkatesan, cramerj

[-- Attachment #1: Type: text/plain, Size: 955 bytes --]

My configuration is two Dell Optiplex PCs ( /proc/cpuinfo attached)
connected back to back.. I am using the vanilla 2.6.14.3 kernel, and
using the e1000 driver for the onboard Intel PRO 1000 Ethernet Card on
both machines. I am performing  both Full and Half Duplex Data
Transfer  and the figures which I am getting are abysmal..

Full Duplex Data Transfer ( Both send and receive )
Ettcp Send  - 55167.45 KB/sec
Ettcp Receive - 54902.49 KB/sec
Total Throughput ~ 110MB/sec
CPU IDLE - 55 percent

Half Duplex Data Transfer ( Sending from one host to another )
Total Throughput = 103 MB/Sec
CPU IDLE - 65 percent

Now, I assume that on Gigabit ethernet, I should be getting Line Rate,
which is around 220 MBps. Even the CPU is not getting max-ed out here
and I am at a loss to understand this behaviour.

Please find the ethtool dump, and the ettcp send and receive dumps for
both full and half duplex transfers

Regards,
Ashutosh

[-- Attachment #2: ettcp-recv-full-duplex.txt --]
[-- Type: text/plain, Size: 618 bytes --]

[root@kir9060 root]# ettcp -s -r
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
nttcp-r: socket
ttcp-r: accept from 192.168.80.15
User: 2249  Nice: 58  System 1694  Idle:111197  Params:4
User: 2278  Nice: 58  System 2250  Idle:114639  Params:4
User_diff: 29  Nice_diff: 0  System_diff: 556  Idle_diff:3442 Tot:4027
User: 0.720139  System: 13.806804  Nice 0.000000  Idle:85.473057
nttcp-r: Buflen:8192 SysLoad:13.81% 3373203456 bytes 60.00secs =54902.49 KB/sec
nttcp-r: 627033 I/O calls, msec/call = 0.10, calls/sec = 10450.57
nttcp-r: 0.2user 6.9sys 1:00real 12% 0i+0d 0maxrss 0+15pf 448574+141csw






[-- Attachment #3: ettcp-send-full-duplex.txt --]
[-- Type: text/plain, Size: 642 bytes --]

[root@kir9060 root]# ettcp -s -t -i 60 -l 65536 192.168.80.15
ttcp-t: buflen=65536, nbuf=2048, align=16384/0, port=5001  tcp  -> 192.168.80.15nttcp-t: socket
nttcp-t: connect
User: 2249  Nice: 58  System 1673  Idle:110852  Params:4
User: 2275  Nice: 58  System 2206  Idle:114417  Params:4
User_diff: 26  Nice_diff: 0  System_diff: 533  Idle_diff:3565 Tot:4124
User: 0.630456  System: 12.924345  Nice 0.000000  Idle:86.445199
nttcp-t: Buflen:65536 SysLoad:12.92% 3389521920 bytes 60.00secs =55167.45 KB/secnttcp-t: 51720 I/O calls, msec/call = 1.19, calls/sec = 861.99
nttcp-t: 0.0user 2.5sys 1:00real 4% 0i+0d 0maxrss 0+27pf 51415+35csw






[-- Attachment #4: ettcp-half-dup.txt --]
[-- Type: text/plain, Size: 639 bytes --]

[root@kir9060 root]# ettcp -s -t -i 60 -l 65536 192.168.80.15
ttcp-t: buflen=65536, nbuf=2048, align=16384/0, port=5001  tcp  -> 192.168.80.15nttcp-t: socket
nttcp-t: connect
User: 2337  Nice: 58  System 2590  Idle:134970  Params:4
User: 2344  Nice: 58  System 2906  Idle:139352  Params:4
User_diff: 7  Nice_diff: 0  System_diff: 316  Idle_diff:4382 Tot:4705
User: 0.148778  System: 6.716259  Nice 0.000000  Idle:93.134963
nttcp-t: Buflen:65536 SysLoad:6.72% 6341132288 bytes 60.00secs =103207.73 KB/secnttcp-t: 96758 I/O calls, msec/call = 0.63, calls/sec = 1612.62
nttcp-t: 0.0user 3.2sys 1:00real 5% 0i+0d 0maxrss 0+27pf 96136+2csw





[-- Attachment #5: cpuinfo.txt --]
[-- Type: text/plain, Size: 439 bytes --]

processor	: 0
vendor_id	: GenuineIntel
cpu family	: 15
model		: 2
model name	: Intel(R) Pentium(R) 4 CPU 2.80GHz
stepping	: 9
cpu MHz		: 2793.594
cache size	: 512 KB
fdiv_bug	: no
hlt_bug		: no
f00f_bug	: no
coma_bug	: no
fpu		: yes
fpu_exception	: yes
cpuid level	: 2
wp		: yes
flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr
bogomips	: 5593.94




[-- Attachment #6: ethtool.txt --]
[-- Type: text/plain, Size: 605 bytes --]

Settings for eth0:
	Supported ports: [ TP ]
	Supported link modes:   10baseT/Half 10baseT/Full 
	                        100baseT/Half 100baseT/Full 
	                        1000baseT/Full 
	Supports auto-negotiation: Yes
	Advertised link modes:  10baseT/Half 10baseT/Full 
	                        100baseT/Half 100baseT/Full 
	                        1000baseT/Full 
	Advertised auto-negotiation: Yes
	Speed: 1000Mb/s
	Duplex: Full
	Port: Twisted Pair
	PHYAD: 0
	Transceiver: internal
	Auto-negotiation: on
	Supports Wake-on: umbg
	Wake-on: g
	Current message level: 0x00000007 (7)
	Link detected: yes

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [RFC] Poor Network Performance with e1000 on 2.6.14.3
       [not found] ` <43E04712.6080108@candelatech.com>
@ 2006-02-01 12:19   ` Ashutosh Naik
  0 siblings, 0 replies; 2+ messages in thread
From: Ashutosh Naik @ 2006-02-01 12:19 UTC (permalink / raw)
  To: Ben Greear, linux-kernel, netdev, linux-net, Andrew Morton, jgarzik

[-- Attachment #1: Type: text/plain, Size: 821 bytes --]

On 2/1/06, Ben Greear <greearb@candelatech.com> wrote:
> Ashutosh Naik wrote:
> > Now, I assume that on Gigabit ethernet, I should be getting Line Rate,
> > which is around 220 MBps. Even the CPU is not getting max-ed out here
> > and I am at a loss to understand this behaviour.
>
> Make sure you are running 64-bit 100+Mhz PCI, otherwise you will not
> get line speed.  ethtool -d [device]
> may get that info for you.

Thanks for that Ben. I now used a 64 bit,133 MHz PCI-X on both
machines and I got around  180 MBps, (86259.48 + 95164.90 ), which is
still a decent way away from Line speed. I think PCI is not a
bottleneck now, although I could be wrong. What could I be missing,
and has anybody seen line speed ( 220 MBps ) with e1000 ?

Regards and Thanks
Ashutosh

ps - Attaching my ettcp log

[-- Attachment #2: log.txt --]
[-- Type: text/plain, Size: 1224 bytes --]

[root@localhost ~]# ettcp -s -t -i 300ttcp-r: accept from 192.168.90.100
User: 17553  Nice: 0  System 8885  Idle:737051  Params:4

ttcp-t: buflen=65536, nbuf=2048, align=16384/0, port=5001  tcp  -> 192.168.90.10
0
nttcp-t: socket
nttcp-t: connect
User: 17555  Nice: 0  System 8892  Idle:737076  Params:4
^[
User: 19818  Nice: 0  System 14716  Idle:747604  Params:4
User_diff: 2265  Nice_diff: 0  System_diff: 5831  Idle_diff:10553 Tot:18649
User: 12.145423  System: 31.267092  Nice 0.000000  Idle:56.587485        
nttcp-r: Buflen:65536 SysLoad:31.27% 29232070656 bytes 299.97secs =95164.90 KB/s
ec
nttcp-r: 699884 I/O calls, msec/call = 0.44, calls/sec = 2333.15
nttcp-r: 0.2user 32.6sys 4:59real 10% 0i+0d 0maxrss 0+30pf 691228+3947csw
User: 19834  Nice: 0  System 14722  Idle:747623  Params:4
User_diff: 2279  Nice_diff: 0  System_diff: 5830  Idle_diff:10547 Tot:18656
User: 12.215909  System: 31.250000  Nice 0.000000  Idle:56.534091        
nttcp-t: Buflen:65536 SysLoad:31.25% 26499153920 bytes 300.00secs =86259.48 KB/s
ec
nttcp-t: 404345 I/O calls, msec/call = 0.76, calls/sec = 1347.80
nttcp-t: 0.2user 44.9sys 5:00real 15% 0i+0d 0maxrss 0+30pf 374765+1306csw
[1]+  Done                    ettcp -s -r -l 65536





^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2006-02-01 12:19 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2006-02-01  5:17 [RFC] Poor Network Performance with e1000 on 2.6.14.3 Ashutosh Naik
     [not found] ` <43E04712.6080108@candelatech.com>
2006-02-01 12:19   ` Ashutosh Naik

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).