Re: Throughput question on Linux system
From: David E. Konerding DSD staff (dek_at_bosshog.lbl.gov)
Date: 08/08/05
- Previous message: shiva: "Re: how to know the exact place where kerenl has crashed"
- In reply to: GS: "Throughput question on Linux system"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Mon, 8 Aug 2005 17:55:42 +0000 (UTC)
In article <1123515017.699960.234690@g44g2000cwa.googlegroups.com>, GS wrote:
> We connected two smart bits ports to two Ethernet ports on the Linux
> system, Smartbits pumps traffic on one port and receives traffic on
> other port, we are getting higher through with lower packet size, and
> getting lower throughput with higher packet size. If the packet size is
> smaller, I should get more packets and which interrupts more and I
> should get lower throughput, Does this some issue with Bottom halfs in
> my Network driver?. Any idea?. Thanks.
The assumption that smaller, more frequent packets lead to greater number
of interrupts can be false if the driver (common only in very high speed cards)
uses interrupt coalescence. Also, there are a number of other assumptions you
make about where the bottleneck is that affect your interpretation of your observtions.
Dave
- Previous message: shiva: "Re: how to know the exact place where kerenl has crashed"
- In reply to: GS: "Throughput question on Linux system"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
