100Mbps to 1000Mbps Performance issue

[inghau]

Hello everybody

i 've just finished setting up a new L3 environment in my costumer's LAN, previously they're using a 2610 router for their interVLAN routing

they have 3 VLAN (VLAN 1/Legacy, VLAN 21/SAP_Server's farm and VLAN 22/SAP_Client)

All the SAP Server's are newest IBM Server's and the spec is about 2-4 processor of Pentium3 Xeon with more than 1Gb of RAM and using "Broadcom 10/100/1000T Ethernet cards"

before upgrading to the new L3 we have few benchmark test using "iperf" tool to see the throughput traffic performance in and between VLAN's and the results are just like :

1. between the same VLAN the throughput is about 80-90Mbps

2. from VLAN 21 to VLAN 22 the throughput is ONLY about 7-10 Mbps (i assume this is happening because we're using 2610 to route between VLAN instead of using L3 switches)

after they have the budget to upgrade their network so i proposed a new L3 Switche (Cat 3550 12T) to replace the R 2610 for the InterVLAN switching

and after configure it today we run the test again just like before upgrading.

The new environment i've just set up is like this

- We have upgraded the NIC on the server's side from 10/100Mbps to 1000Mbps and directly connected to the 3550-12T switches

- the client's still using 10/100 Mbps and connected to other 3500XL-12 and some 3500XL-24

- The link between each switch is using 1000SX and some using 1000BaseT

And the test results is : (all the test are running on the test environment (i mean is there is no such user activity, the switch and server is just performing the bench)

1 Between the same VLAN 22 (10/100Mbps Environment) the throughput is about 80-90Mbps

2 Between VLAN 21 and 22 the throughput is just about the same 80-90Mbps

(this is as i've been expecting BUT )

3 BUT between the same VLAN 21 (FULL 1000BaseT environment and directly connected to the 3550-12T) the performance is just 160-200Mbps only

Actually this is my first real L3 configuration using L3 Switches. and luckyly i can configure it to work (thanks to the config example)

My question is :

WHY the performance in the full 1000BaseT environment is just only 160-200Mbps (only about 2 times than the 100Mbps environment), i assume the improvement should be about 500-600Mbps, please correct me if i'm wrong.

If some of you would like to see my configuration on the L3 switches please just let me know i'll be happy to post it here.

Thank you very much for any comments

Best Regards

@ I n g - H a u @

[wong34539]

Be sure that you are hard coding the speed and duplex.

[inghau]

thank you for your suggest.

i've tried to put the speed 1000 and full-duplex command but the bench result is still the same.

how about your experiance in 1000baseT servers what's your highest throughput ?

can u share it with me ?

[ns]

Are you monitoring the servers e.g. the CPU, NIC, disk etc with perfmon?

Look for a bottleneck there.

But just in case are you monitoring the 3550T e.g. CPU?

What NIC's are you using? Driver issue? If they are not IBM or other OEM cards, change them.

[inghau]

I'm only monitoring the CPU processor using the task manager tool (the server is 2000 advanced server)

the cpu usage is just below 20% when i perform the test.

but i havent checked the disk/nic using perfmon. what kind of values to check there ?

the NIC is from IBM i forgot the type but the driver was also the newest one, because i see the sys engineer from IBM who performing the installation.

is there any possibilities that the benchmark tool i'm using (iperf) didn't support gigabit ethernet ?

thanks for any comments and suggest

Ing

[ns]

bytes transferred/sec,and any parameter that might indicate a bottleneck. Also what "version" of the SCSI protocol are you using? Ultra2, Ultra160 or Ultra320 etc. make sure that you're not maxing out the SCSI bus.

I'm not familiar with the iperf tool, sorry.

Cheers.

[glen.grant]

What you are seeing could be correct , just read about a test using full duplex fast ethernet versus gigabit and in some cases the transfer rates were not much better for the gigabit versus full duplex fast ether . Maybe the servers can't pump it out any faster than what you are seeing ...

[inghau]

well okay maybe i'll try to find the whitepaper about 100TX vs 1000T

thank you very much everybody