[Beowulf] raw ethernet
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
050675 at student.unife.it 050675 at student.unife.itFri Jul 23 00:04:22 PDT 2004
- Previous message: [Beowulf] About Beowulf in general
- Next message: [Beowulf] raw ethernet
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Hi all, I've seen yours mails in response to my first post about raw ethernet, so I would like to better explain my project. At this moment I'm trying to study raw ethernet performances in a sort of dummy way that means that I made two programs, the first one to send raw packets and the second one to receive them and make statistics (received frames, lost frames, time spent, throughput and so on). My purpose is to study raw ethernet performances and to understand the problems linked to this approach (raw ethernet). For example, I've seen that raw ethernet is unreliable so I loose packets (even if my two nodes are connected by a point-to-point gigabit link). I'm interested in understand why I have a packet loss (interrupt management, full rx ring, rx buffer overflow, each of them combined?) Now I'm trying to modulate the problem with instruments such as ifconfig, or other infos captured by some kernel's constants (i.e. /proc/net/softnet_stat), but I'm searching tools which can give me more precision and accurate results. In your opinion what can I use? In alternative I can see the driver source to understand how and where it allocates rxdescriptors, or sk_bufs to storage incoming packets even if it seems a long work. Simone Saravalli
- Previous message: [Beowulf] About Beowulf in general
- Next message: [Beowulf] raw ethernet
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
