嗯嗯 OK 我回去再PO成績給大家看看
打算來敗一張Intel PCIE 的Gigabit網卡給桌機 然後再買一張Intel PCI的Gigabit網卡給準系統
然後再買一張VIA的便宜Gigabit網卡來交叉測試

家中測試
兩邊一樣是intel pro/1000 GT
中間串兩個giga switch (lantech/ZyXEL)
五年前佈的Cat.5E線路總長超越30米 (樓上啪到樓下那段是直接露天明線)
結果是可憐的381M
原因是整個cabling system太鳥了

是的, 執行sync可將還在記憶體中的資料寫回硬碟
而小弟指的是在/etc/fstab掛載磁區時, 下的參數
如果有下sync, 則磁碟與記憶體中的資料會以同步寫入的方式動作(系統預設為async),

檢查方法為, # mount 查看就可以了

我檢查了之後，發現都沒有說，底下的情況是怎樣阿
root@justin-desktop:/home/justin# mount
/dev/hda1 on / type ext3 (rw,errors=remount-ro)
proc on /proc type proc (rw)
/sys on /sys type sysfs (rw)
varrun on /var/run type tmpfs (rw)
varlock on /var/lock type tmpfs (rw)
procbususb on /proc/bus/usb type usbfs (rw)
udev on /dev type tmpfs (rw)
devpts on /dev/pts type devpts (rw,gid=5,mode=620)
devshm on /dev/shm type tmpfs (rw)
lrm on /lib/modules/2.6.15-26-386/volatile type tmpfs (rw)

我的更濫，桌機(內建Marvel 88E8053)傳到準系統(螃蟹8169s)
只有347Mbits/sec，唉!真的是該換了，真是濫到爆的Gigabit級網路

剛剛用了小弟兩年前買的筆電內附的Broadcom Gigabit網卡，更新驅動前的成績是570Mbit/s
更新驅動後的成績是 660Mbit/s，真的是讓我...
桌機跟額外買的網卡竟然比不上筆電內附的><

單純一點用封包測吧，提供一個小軟體。

鵝來貼個linux下的測試結果好了:p...BTW,測試環境是以約2m的CAT5e直接對接,client是S754的Sempron2500+1GB DDR400+PCI-E 88E8053,server是AM2的Sempron3000+2GB DDRII667+MCP51內建的GbE(PHY也是Marvell的),沒開Jumbo Frame(因為linux不知為何沒辦法開off loading及Jumbo Frame等功能:ase)....

只開一個connection,88E8053 default有開offloading時

root@lab01:~# ethtool -k eth0
Offload parameters for eth0:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp segmentation offload: on
root@lab01:~# time iperf -c 192.168.1.115
------------------------------------------------------------
Client connecting to 192.168.1.115, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[ 5] local 192.168.1.110 port 56285 connected with 192.168.1.115 port 5001
[ 5] 0.0-10.0 sec 1.09 GBytes 938 Mbits/sec

real 0m10.035s
user 0m0.064s
sys 0m2.008s

throughput約938Mbps,CPU loading約20%....

同上,關閉所有off loading....

root@lab01:~# ethtool -k eth0
Offload parameters for eth0:
rx-checksumming: off
tx-checksumming: off
scatter-gather: off
tcp segmentation offload: off
root@lab01:~# time iperf -c 192.168.1.115
------------------------------------------------------------
Client connecting to 192.168.1.115, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[ 5] local 192.168.1.110 port 56286 connected with 192.168.1.115 port 5001
[ 5] 0.0-10.0 sec 998 MBytes 837 Mbits/sec

real 0m10.020s
user 0m0.044s
sys 0m3.308s

throughput下降約10%,CPU loading由20%上升到33%(約60%)....

重新打開所有off loading,開10個connection....

root@lab01:~# ethtool -k eth0
Offload parameters for eth0:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp segmentation offload: on
root@lab01:~# time iperf -c 192.168.1.115 -P10
------------------------------------------------------------
Client connecting to 192.168.1.115, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[ 5] local 192.168.1.110 port 36511 connected with 192.168.1.115 port 5001
[ 6] local 192.168.1.110 port 36512 connected with 192.168.1.115 port 5001
[ 7] local 192.168.1.110 port 36513 connected with 192.168.1.115 port 5001
[ 8] local 192.168.1.110 port 36514 connected with 192.168.1.115 port 5001
[ 9] local 192.168.1.110 port 36515 connected with 192.168.1.115 port 5001
[ 14] local 192.168.1.110 port 36520 connected with 192.168.1.115 port 5001
[ 10] local 192.168.1.110 port 36516 connected with 192.168.1.115 port 5001
[ 11] local 192.168.1.110 port 36517 connected with 192.168.1.115 port 5001
[ 12] local 192.168.1.110 port 36518 connected with 192.168.1.115 port 5001
[ 13] local 192.168.1.110 port 36519 connected with 192.168.1.115 port 5001
[ 5] 0.0-10.0 sec 113 MBytes 94.8 Mbits/sec
[ 6] 0.0-10.0 sec 113 MBytes 94.7 Mbits/sec
[ 7] 0.0-10.0 sec 113 MBytes 94.7 Mbits/sec
[ 9] 0.0-10.0 sec 112 MBytes 94.3 Mbits/sec
[ 10] 0.0-10.0 sec 112 MBytes 93.7 Mbits/sec
[ 8] 0.0-10.0 sec 113 MBytes 94.6 Mbits/sec
[ 14] 0.0-10.0 sec 112 MBytes 94.0 Mbits/sec
[ 11] 0.0-10.0 sec 112 MBytes 93.9 Mbits/sec
[ 13] 0.0-10.0 sec 112 MBytes 93.8 Mbits/sec
[ 12] 0.0-10.0 sec 112 MBytes 93.8 Mbits/sec
[SUM] 0.0-10.0 sec 1.10 GBytes 939 Mbits/sec

real 0m10.086s
user 0m0.080s
sys 0m2.044s

CPU loading及throughput都與開一個connection時約略相當....

同上,關閉所有off loading....

root@lab01:~# ethtool -k eth0
Offload parameters for eth0:
rx-checksumming: off
tx-checksumming: off
scatter-gather: off
tcp segmentation offload: off
root@lab01:~# time iperf -c 192.168.1.115 -P10
------------------------------------------------------------
Client connecting to 192.168.1.115, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[ 14] local 192.168.1.110 port 58168 connected with 192.168.1.115 port 5001
[ 5] local 192.168.1.110 port 58159 connected with 192.168.1.115 port 5001
[ 6] local 192.168.1.110 port 58160 connected with 192.168.1.115 port 5001
[ 7] local 192.168.1.110 port 58161 connected with 192.168.1.115 port 5001
[ 8] local 192.168.1.110 port 58162 connected with 192.168.1.115 port 5001
[ 9] local 192.168.1.110 port 58163 connected with 192.168.1.115 port 5001
[ 10] local 192.168.1.110 port 58164 connected with 192.168.1.115 port 5001
[ 11] local 192.168.1.110 port 58165 connected with 192.168.1.115 port 5001
[ 12] local 192.168.1.110 port 58166 connected with 192.168.1.115 port 5001
[ 13] local 192.168.1.110 port 58167 connected with 192.168.1.115 port 5001
[ 6] 0.0- 7.0 sec 71.8 MBytes 86.0 Mbits/sec
[ 9] 0.0- 7.0 sec 71.3 MBytes 85.4 Mbits/sec
[ 11] 0.0- 7.0 sec 69.8 MBytes 83.7 Mbits/sec
[ 12] 0.0- 7.0 sec 69.8 MBytes 83.6 Mbits/sec
[ 5] 0.0- 7.0 sec 71.5 MBytes 85.6 Mbits/sec
[ 7] 0.0- 7.0 sec 71.7 MBytes 85.8 Mbits/sec
[ 8] 0.0- 7.0 sec 70.2 MBytes 83.9 Mbits/sec
[ 10] 0.0- 7.0 sec 71.5 MBytes 85.5 Mbits/sec
[ 13] 0.0- 7.0 sec 71.1 MBytes 85.1 Mbits/sec
[ 14] 0.0-10.0 sec 372 MBytes 312 Mbits/sec
[SUM] 0.0-10.0 sec 1010 MBytes 847 Mbits/sec

real 0m13.027s
user 0m0.068s
sys 0m3.960s

throughput約略與只開一個connection關off loading時相當,CPU loading由33%上升至40%(約20%)....

總結看來還不錯(吃CPU沒想像中的兇:p)....BTW,鵝發現iperf在M$ Win中開的buffer(TCP win)有點小---default只有8k,不知是否是iperf win32版binary版本較舊所致,若調大時(i.e. -w 64k)對throughput有不小的助益,測試結果不理想者可以試試:p:p....

奇怪了，為什麼鵝兄你的Marvel 88E8053就跑的那麼好
我的就這麼差呢?? 我的總輸出只有到347Mbit/sec
然後我那台linux的準系統插的螃蟹卡值也只能到350Mbit/sec左右
linux需要調整什麼才可以讓速度快一點嗎？我的桌機傳到準系統的linux
只有14MB左右的速度，但是我從準系統下載卻有27MB的速度，為什麼會不一樣呢？

鵝前面提過win32版iperf的buffer(TCP window) default只有8KB,設成64KB之後throughput就好多了,除此之外鵝沒做啥特別的調校啊:confused: ....BTW,鵝用的kernel是重新compile過的v2.6.17,也先把iptable的rule清掉了,您要不要檢查一下是不是iptable的關係(一般常見的distribution如RH/FC之類會有default的rule,iptable/firewall rule的寫法對throughput/loading會有重大的影響:p).....



這就牽涉到真正的I/O和file system了,光由您提到有限的資訊恐怕是很難做出有意義的判斷的:p:p....