cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
361
Views
0
Helpful
4
Replies

Switch throughput..how much to expect.

rajitoor55
Level 1
Level 1

While troubleshooting a slowness issue, I connected 2 laptops directly to a 3750-48 port switch. One made a http web file server and folder shared and started downloads and file copy on the other. Download rate that i got was of 18MBytes(144Mbits/sec). To test further i used Jperf and rate with this default TCP settings was 44MBytes(352 Mbits/sec). At the time of testing both ports are in auto full duplex, with a speed of 1000. And switch itself is not having considerable load over other interfaces, only few multiples of 100Kbits. Switch CPU utilization is low to 7-8%. On 1Gig interface transfer rate not even reaches half of what it supports     

1 Accepted Solution

Accepted Solutions

Peter Paluch
Cisco Employee
Cisco Employee

Hello,

Doing reliable performance tests is not that trivial. Your own laptops may be the bottlenecks. I would first suggest to connect them together back-to-back using a crossover cable with no switch inbetween, and make sure you can repeatedly and sustainably get a high performance out of the testing setup you're using. Keep in mind that to test 1Gbps link close to its limits (125 MByte/s), the hard drives on both laptops would need to provide a sustained read/write performance of at least 100 MByte/s. This may not always be the case with ordinary laptop hard drives. Using RAM-emulated disks (and having enough memory to prevent swapping which would revert to the performance limits imposed by hard drives) is highly recommended. In addition, the operating system should be running as few processes as possible, and ideally, the server and client processes should be running with elevated scheduling priorities. Can you confirm that with these two laptops connected via a direct cross cable, you are repeatedly and sustainably getting transmission speeds close to 1Gbps?

Plain Layer2 switching is never handled by the CPU in switches, therefore, if you are testing the throughput in a single VLAN, the CPU load is irrelevant - and even then, you have indicated yourself it is negligible.

Best regards,
Peter

View solution in original post

4 Replies 4

Peter Paluch
Cisco Employee
Cisco Employee

Hello,

Doing reliable performance tests is not that trivial. Your own laptops may be the bottlenecks. I would first suggest to connect them together back-to-back using a crossover cable with no switch inbetween, and make sure you can repeatedly and sustainably get a high performance out of the testing setup you're using. Keep in mind that to test 1Gbps link close to its limits (125 MByte/s), the hard drives on both laptops would need to provide a sustained read/write performance of at least 100 MByte/s. This may not always be the case with ordinary laptop hard drives. Using RAM-emulated disks (and having enough memory to prevent swapping which would revert to the performance limits imposed by hard drives) is highly recommended. In addition, the operating system should be running as few processes as possible, and ideally, the server and client processes should be running with elevated scheduling priorities. Can you confirm that with these two laptops connected via a direct cross cable, you are repeatedly and sustainably getting transmission speeds close to 1Gbps?

Plain Layer2 switching is never handled by the CPU in switches, therefore, if you are testing the throughput in a single VLAN, the CPU load is irrelevant - and even then, you have indicated yourself it is negligible.

Best regards,
Peter

Joseph W. Doherty
Hall of Fame
Hall of Fame

Do the interfaces show drops?  If so, do drops increment during transfer testing?

Is QoS enabled?
 

No drops..no QoS

rajitoor55
Level 1
Level 1

Thanks Peter, Yes it indeed is the laptop throughput that is throttling. 

Review Cisco Networking products for a $25 gift card