I was using the TimingSimple CPU. Ok, I'll try the O3 model and a higher frequency and check.
Thanks, Pritha On Mon, Apr 23, 2012 at 4:52 PM, Steve Reinhardt <ste...@gmail.com> wrote: > Are you using the TimingSimple CPU or the O3 model? The latter will give > you faster effective performance at the cost of significantly increasing > simulation time. > > In either case, you may want to try just cranking up the CPU frequency. > > Steve > > > On Mon, Apr 23, 2012 at 2:42 PM, Pritha Ghoshal <pritha9...@tamu.edu>wrote: > >> I am modelling an Alpha CPU, timing mode, with the default frequency. >> >> Pritha >> >> On Mon, Apr 23, 2012 at 4:39 PM, Steve Reinhardt <ste...@gmail.com>wrote: >> >>> What sort of CPU are you modeling? If your CPU is configured to be too >>> slow, it won't be able to keep up with the packets. >>> >>> On Mon, Apr 23, 2012 at 12:52 PM, Pritha Ghoshal <pritha9...@tamu.edu>wrote: >>> >>>> Hi, >>>> >>>> I am working with the i8254xgbe driver. I am trying to run a network >>>> trace file which has maximum bandwidth of 10Gbps through this NIC(average >>>> BW seems to be around 3-4Gbps). I am calling the receive function of eth0 >>>> the trace and rerouting it back through eth1. But I am seeing a lot of >>>> packet drops on eth0(around 98-99% packets are getting dropped). The >>>> txBandwidth seems to be around 3-4Mbps and therefore the tail of the >>>> descriptor ring in host memory - RDT was not being updated soon enough for >>>> all the packets in FIFO to be copied over to the main memory. I did the >>>> following things to try to improve it and the outcomes: >>>> >>>> 1. Increase the rxFifo size. -> Delayed the beginning of packet drop >>>> 2. Increased the bus speeds -> Delayed the beginning of packet drop >>>> 3. Decreased number of RXDMT interrupts (I think there is a bug in the >>>> code which causes these interrupts where the number of descriptors was >>>> being compared to the sizeof the descriptors in bytes) -> Did not help >>>> probably. >>>> 4. Modified the number of descriptors in the circular buffer using >>>> ethtool from 256(default) to 4096(maximum) -> Delayed the beginning of >>>> packet drop >>>> 5. Modified the NAPI weight in the kernel code in >>>> /drivers/net/e1000/e1000_main.c to 8 -> Improved the txBandWidth to around >>>> 30Mbps in timing mode, 3-4Gbps in atomic mode. But I need around 4-5Gbps in >>>> timing mode. >>>> 6. Finally was trying to install a later version of the e1000 driver >>>> (the present version is 7.3.21-k3-NAPI). I tried to download 8.0.16 >>>> version and copy the files over in place of the e1000 files present in the >>>> linux kernel directory and built it. But this does not seem to work, the >>>> eth interfaces are not coming up during boot and if I try modprobe e1000 it >>>> gives: >>>> # modprobe e1000 >>>> FATAL: Could not load /lib/modules/2.6.27.61-dirty/modules.dep: No such >>>> file or directory >>>> >>>> Does anyone have any suggestions regarding this about how to reduce >>>> packet drops? >>>> >>>> Thanks, >>>> Pritha >>>> >>>> _______________________________________________ >>>> gem5-users mailing list >>>> gem5-users@gem5.org >>>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >>>> >>> >>> >>> _______________________________________________ >>> gem5-users mailing list >>> gem5-users@gem5.org >>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >>> >> >> >> _______________________________________________ >> gem5-users mailing list >> gem5-users@gem5.org >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> > > > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
