Content-type: multipart/alternative; boundary="Boundary_(ID_1eEfFOw70a+0zIVNHnxshQ)" --Boundary_(ID_1eEfFOw70a+0zIVNHnxshQ) Content-type: text/plain; charset=us-ascii Content-transfer-encoding: 7BIT Greeting, Recently, I am working on low level optimizations related to network performance, and get a question regarding to Intel Xeon processor. Could anybody help me out? Thanks in advance. The question is this: We use iperf to run data transmission experiments from a sender to a receiver. The sender has two Intel Xeon 5150 processors. Except iperf, no other applications run in the sender. Also, in the sender, we pin "iperf" and "traffic interrupts" to one specific core (say core 1). Then we run oprofile to profile the system. Since we pin "iperf" and "traffic interrupts" to one specific core (say core 1), we had thought that there should be no much SNOOP traffic in the system. However, the profiling results show lots of HITM traffic, which is to our shock! The cache content should not be updated in our cores' cache, where does these HITM traffic come from?! Or anything wrong with oprofile? I have attached a detailed description to the email. Really appreciate if you could help me out. Yours sincerely, Wenji --Boundary_(ID_1eEfFOw70a+0zIVNHnxshQ) Content-type: text/html; charset=us-ascii Content-transfer-encoding: 7BIT

Greeting,

 

Recently, I am working on low level optimizations related to network performance, and get a question regarding to Intel Xeon processor. Could anybody help me out? Thanks in advance.

 

 

The question is this:

 

We use iperf to run data transmission experiments from a sender to a receiver. The sender has two Intel Xeon 5150 processors. Except iperf, no other applications run in the sender. Also, in the sender, we pin “iperf” and “traffic interrupts” to one specific core (say core 1). Then we run oprofile to profile the system.

 

Since we pin “iperf” and “traffic interrupts” to one specific core (say core 1), we had thought that there should be no much SNOOP traffic in the system. However, the profiling results show lots of HITM traffic, which is to our shock! The cache content should not be updated in our cores’ cache, where does these HITM traffic come from?! Or anything wrong with oprofile?

 

I have attached a detailed description to the email.

 

Really appreciate if you could help me out.

 

Yours sincerely,

 

Wenji

 

--Boundary_(ID_1eEfFOw70a+0zIVNHnxshQ)--