Dear Professor Hyesoon,
I have reduced the cache size, but the network injection rate is still low. I use the SMT4 core, and each core has four single-thread applications. I think the reason is that when one thread is waiting the miss data/inst, the thread will be blocked. So when one thread is waiting for the miss data/inst, this thread will not send another packet to the network. Is this right?
I checked the KONB "num_inst_to_fetch_after_load" in the source code, and found it didn't work. So I don't know how many instructions can be fetched and executed after load miss.
Is there some parameters which about the LSQ(load store queue), or LMQ(load miss queue) or ISQ(Issure queue)?
And what is the difference between the KNOB"large_width" and "schedule_to_large_width"?
Thanks for your time!!
Best regards,
Applee
在 2016年4月2日星期六 UTC+8下午8:59:43,Hyesoon写道: