Question on throughput

- P
- pfisterfarm
  
  Contact options for registered users
posted
15 years ago

Wed, Apr 1, 2009 8:19 PM

I was trying out the ttcp command to get throughput numbers (I've just now found out about it...). I probably should know this, but it's been a while since I've read anything about throughput -- what is an acceptable value for throughput? For example, I ran a test between my PC and the Cisco switch it's directly connected to (100mbps connection, full duplex). It came back at around 17mbps. Is this normal, or should it be higher? I know it will be a fair bit less than the theoretical maximum, but I wasn't sure how much less. I don't see any errors on the port.

Thanks!

- T
- Trendkill
  
  Contact options for registered users
Vote on answer
posted
15 years ago

Wed, Apr 1, 2009 8:55 PM

You should easily get 12MBps, which equates to 96mbps, obviously depending on the nature of the application, the end points (both source and destination), and hard disk/IO. FTP on most any box today should run upwards of 12MBps (presuming your switch is semi current), and if you had a gig switch and gig network card, a few hundred Mbps. It generally takes server class gear to begin pushing beyond that from my experience, but today's dual cores with 10k drives could easily cannibalize a gig link (or a similar percentage of gig as the 100 megabyte example). As for multi-gig, I've only seen enterprise class hardware push in the multi-gigs sustainably through etherchannel.

- B
- bod43
  
  Contact options for registered users
Vote on answer
posted
15 years ago

Wed, Apr 1, 2009 11:22 PM

OP seems to say that one end point is a switch of some kind. The switch CPU could well be the limiting point.

You could check the CPU during the test to see if that was limiting the throughput.

Modern PCs (say 2GHz plus) will pretty much saturate a GBE, I seem to recall. Its easy to figure out, just check the CPU load.

One nice thing about TTCP is that it does not invoke the disk and so factors one potential variable out of the equation:)

A *very* nice thing to do with ttcp is to run more than one copy of it. Add copies until the total throughput stops increasing. You then know that ttcp itself is not limiting the performance. This is for sure an issue with high latency links.

Network performance testing is not straightforward.