Re: Upgrade to gigabyte NIC?
From: Susan Bradley, CPA aka Ebitz - SBS Rocks [MVP] (sbradcpa_at_pacbell.net)
Date: 02/21/05
- Next message: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Previous message: Gary Karasik: "Re: Dogs Breakfast, disaster, or just me over-reacting?"
- In reply to: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Next in thread: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Reply: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Messages sorted by: [ date ] [ thread ]
Date: Sun, 20 Feb 2005 19:39:46 -0800
When I've changed a physical NIC [especially the inner one] you have to
do the "loopback" install so that the inner NIC gunk stays with the
server [does that make sense?]
Henry Craven [SBS-MVP] wrote:
> Disable it prior, or the system will complain that you have the same IP
> Assigned twice. to different devices.
>
> I'd be interested to hear if you get the performance gain you expect.
> and whether it's a value ROI.
>
> Just an FYI ------
>
> Fast Ethernet based communication systems behave poorly when sending and
> receiving small packets of data. When using small packets,a temporal
> bottleneck occurs in the communication, causing low throughputs. By
> eliminating some of the possible causes for the bottleneck we were able
> to focus on the CPU as the main hardware reason for it, as the tests
> showed very high and increasing CPU loads when using decreasing size
> packets. By thoroughly testing the possible reasons for that CPU load,
> we came to a conclusion that the main cause of the bottleneck is the
> well known Interrupt Livelock situation, where mass of incoming packets
> cause a very large number of NIC interrupts to the CPU, causing the CPU
> to use almost all of its resources for dealing with the interrupts and
> doing hardly anything else. As the NICs hardware improves and
> communication speed rises to 1GB and more, the pcibus becomes the main
> reason for the communication bottleneck, but as that bottleneck got a
> solution by using a larger MTU size (9KB instead of the regular 1.5KB),
> the interrupt livelock problem still exists and again becomes the
> bottleneck cause, resulting in a much desired solution that will handle
> the livelock and decrease latency to minimum.
>
> and from another article.
>
> ....The disappointing performance we saw with Gigabit Ethernet is
> probably due to both Microsoft's TCP/IP stack and D-Link's more consumer
> level hardware.
> ---------
>
> Keep us posted.
>
-- An open letter to the Security Community:: http://msmvps.com/bradley/archive/2004/12/12/23540.aspx
- Next message: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Previous message: Gary Karasik: "Re: Dogs Breakfast, disaster, or just me over-reacting?"
- In reply to: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Next in thread: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Reply: Henry Craven [SBS-MVP]: "Re: Upgrade to gigabyte NIC?"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
|