I wanted to start a discussion about the actual sustained payload throughput of Fast Ethernet. Personally, I have been unable to clock anything faster than 4.5Mbytes / sec between two full duplex 100Mbit machines on my network. I have Catalyst 6500 switch and have been doing several experiments to try and get higher throughput.
How long does it take to perform a DOS based ftp of 270MBytes.
I have been unable to break the 60 second barrier.
With a windows based ftp (through the browser) Windows says that it is clocking speeds up to 7.5 MBytes / sec but the actual transfer time has been slower than with DOS based ftp transfers.
Some things that I have tried:
I have isolated two test machines on their own VLAN even tried directly connecting them via a crossover cable.
I have manually set them to full duplex 100Mbps.
MTU size is set to 1500 and latency is <1ms.
The machines are using Intel Proset Fast Ethernet NIC's.
I have increased the receive buffers in the NIC firmware to the maximum.
Both machines are running Win2K.
I realize that there is some protocol overhead. But my calculations show that the entire TCP/IP stack should take no more than 5%. That puts the theoretical maximum transfer rate at around 11.8 MBytes / sec. (100Mbits / 8 bits in a byte = 12.5 Mbytes / sec * 95% = 11.875)
Can anyone else reproduce my results? If you are getting faster transfer speeds how are you achieving this? I would also like someone to attempt this test with two UNIX / LINIX machines, to see if this is a OS issue.
Check your cabling: If you are using older Cat5 (5e, 6) jumpers or cabling, use/abuse may have pushed them below specification. If they are hand-made cables, check the pinouts and quaility of the termination at each end to ensure compliance with EIA/TIA specification. A weak cable will kill high-speed performance. You may also want to check any other infrastructure components (panels, plates, inserts) to see that they are also well within the specified rating.
For throughput testing, go to the NetIQ website, and get a (free) copy of Qcheck. Qcheck was designed to check things like throughput, latency, etc, and was coded to minimize influence by background processes, OS quirks, etc. It's basically a stripped-down version of Chariot, which we routinely use in the Lab for application-level performance testing.
I have downloaded QCheck and run the throughput test. I average about 93Mbits/sec (which is about what I would expect) The questions still remains however -- why can I only get 4.5 MBytes /sec through an FTP transfer? FTP cannot have that much protocol overhead. Is it Windows2000? And if so, what can be done to tweak the OS to use less overhead?
www.broadbandreports.com (formerly www.dslreports.com - still works) has a throughput test (up / downstream). They also offer a utility called DrTCP that allows you to adjust your MS Windows TCP/IP parameters (windows size, MTU, etc).
The upstream/downstream test will eventually lead you to a screen with recommendations for adjusting your parameters for optimum performance (using the DrTCP utility).
Defragging the hard drive is an excellent suggestion.
Use "msconfig" to set your system to boot with no additional utilities running (like the system tray stuff).
You mention the 100ve NIC - that's usually a built-in (chipset) or Intel's low-end NIC. If the performance is an issue, you may want to step up to one of the server/workstation-class NICs (or if you do IPSEC / VPNs, try their "S" NICs).
What is the CPU / memory configuration of the machine? Do you have any other interface cards installed (PCI) ?
We are pleased to announce availability of Beta software for 16.6.3. 16.6.3 will be the second rebuild on the 16.6 release train targeted towards Catalyst 9500/9400/9300/3850/3650 switching platforms. We are looking for early feedback from custome...