On Jan 27, 2009, at 8:11 PM, Rick Strong wrote: > Hi all and especially Ali, > > I have been working with the webnew benchmark (specweb95-apache) and > have noticed two problems that kill throughput and was wondering if > you > have encountered them or know what is happening > > (1) All requests handled by the server server are coming back 404 > (File) > Not Found. Nate mentioned that you figured out this was due to a bad > mod_specweb99.so. If you know what is going on in this situation, any > help would be great. I've never heard of the webnew benchmark. I would check the apache configuration and make sure the files are where they should be. The 404 error here has nothing to do with simulation and is just a misconfiguration of the workload.
> > (2) Some simulations (a small subset) using a vanilla M5 kernel have > the > server receive a packet that it believes is Out-Of-Order (even > though it > has the right seq number that the server put in its previous ack). The > result is the client sends a Dup ACK. The server does not respond for > 200 ms and the client then times out and re-requests 6 files on 6 > different ports (6 files in total). A more detailed account follows. > The > time delta between 1-2, 2-3, and 3-4 is 10s of microseconds There are three possibilities here: a) A kernel bug b) a device model/driver bug c) a checkpointing bug (as it relates to (b)) What kernel version are you using? Could you put the ethernet trace somewhere so I could look at it? Ali > > 1): server responds to previous request with ack:232 len:431 > 2): client requests new file (GET) with seq:232 len:77 > 3): server responds with TCP Out-Of-Order with ack:232 len:431 > 4): client responds with TCP Dup ACK with seq:309 len:0 > 5): client times out and re-requests files with seq:239 len:77 200ms > later > > *this data is from the ethernet trace > > If you have experienced this problem and/or know of a solution, please > let me know what I can do to avoid this disruptive pause in packet > transmission. > > Best, > -Rick > _______________________________________________ > m5-dev mailing list > [email protected] > http://m5sim.org/mailman/listinfo/m5-dev > _______________________________________________ m5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/m5-dev
