On 4/25/07, <b class="gmail_sendername">Roland Dreier</b> <<a href="mailto:rdreier@cisco.com">rdreier@cisco.com</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
> 1. Direct access in ib_find_pkey will probably heart RC connections<br> > per second rate.<br><br>I think it's probably in the noise. And anyway I don't think the<br>connection rate of ipoib CM is particularly important. And we can
<br>always optimize ib_find_pkey as part of the longer-term plan to get<br>rid of ib_find_cached_pkey.</blockquote><div><br>
rate of connections per second might be an intresting feature in the
context of doing TCP offload eg with SDP, when you want to see how many
connetions can a web server or database establish with clients in unit
of time. With the cache elimination every new connection will consume
two more IB commands (port query and pkey table read).<br>
</div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">> 2. What do you think about OrG's opinion (I'm copying it from the other thread):
<br><br>He seems to be saying that it's OK to introduce a window where things<br>fail spuriously. I disagree.</blockquote><div><br>
Let it be. It makes sense to me to eliminate the cache.<br>
<br>
Or.<br>
</div></div>