Tom's Hardware > Forum > General Networking > Network General Discussions > CPU saturation = slower ethernet speeds?

CPU saturation = slower ethernet speeds?

Forum General Networking : Network General Discussions - CPU saturation = slower ethernet speeds?

Tom's Hardware: Over 1.4 million members in 6 different countries available to answer all your high-tech questions. Sign up now! Its free!
Word :    Username :           
 

Hello. I noticed that if my CPU utilisation reaches a certain a threshold the speed of my VIA Rhine II adapter drops considerably (from +8MB/s to ~4MB/s).. why is that? And if I switch to Firewire or USB2.0 networking can I get better throughput even when my CPU utilisation is high? I checked my DMA settings and it's enabled for all drives. I could use +10MB/s LAN speed.. Any suggestions?

Sponsored Links
Register or log in to remove.

You're probably using PCI as interface, PCI has high interrupt CPU usage so lots will go to waste. You need a PCIe adapter. It could also be bad drivers. Some drivers can use polling as interrupt method, which would lower the CPU utilization caused by interrupts.

Are you using gigabit networking, and how did you test the speeds?

------------------------------ ...man will occasionally stumble over the truth, but usually manages to pick himself up, walk over or around it, and carry on.
Reply to sub mesa
- 0 +

sub mesa wrote :

You're probably using PCI as interface, PCI has high interrupt CPU usage so lots will go to waste. You need a PCIe adapter. It could also be bad drivers. Some drivers can use polling as interrupt method, which would lower the CPU utilization caused by interrupts.

Are you using gigabit networking, and how did you test the speeds?





sub mesa, this is an interesting piece of info. i have two test machine sets; one set with a pcie-pcie gig-e interfaces and one with a pci-pcie interface. i am working on a setup where the pci-pcie connection will lose packets (about 5-10%). where the pcie-pcie connection will have a packet loss of less than 0.00001%. for the pci-pcie conneton, i have tried 3 other pci gig-e cards and they all produce the same packet loss (about 5-10%).

these are controlled tests with no changes other then the pci cards. the only pcie gige cards i currently have are broadcom BCM5755 chipset based. the pci gige cards vary.

anyway, the point of this post. do you know where i could start looking up hard documentation for why the pci cards...umm..."suck" compared to the pcie cards? i need to find some soild justifications why this happens. ive looked around but dont think im searching the right words. do you know any keywords i should try?

Reply to luw
- 0 +

I'm not sure on this subject, but I think TCP/IP itself requires increased CPU usage the faster it transfers. I know on my PCI-X (not PCIe) GbE server-class card it uses the onboard hardware chip to reduce load on CPU.

------------------------------ GTL Ref Tweaking Guide - PM for detail
Brand is for the weak-minded, only product matters.
Resilient to marketing.
http://farm3.static.flickr.com/2520/3850950213_aea6bbbcfe_o.jpg
Reply to wuzy
Tom's Hardware > Forum > General Networking > Network General Discussions > CPU saturation = slower ethernet speeds?
Go to:

There are 748 identified and unidentified users. To see the list of identified users, Click here.

Sponsored links
  • Ask the community now
  • Publish
Ad
They won a badge
Join us in greeting them