slides from dlg & me speaking at cuug meeting, prodded by mcbride

openbsd · May 31, 2007 · c760ff7 · c760ff7
1 parent b123016
commit c760ff7
Show file tree

Hide file tree

Showing 106 changed files with 573 additions and 2 deletions.
diff --git a/papers/cuug2007/index.html b/papers/cuug2007/index.html
@@ -0,0 +1,33 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="mgp00001.html"><IMG SRC="mgp00001.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 1"></A>
+<A HREF="mgp00002.html"><IMG SRC="mgp00002.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 2"></A>
+<A HREF="mgp00003.html"><IMG SRC="mgp00003.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 3"></A>
+<A HREF="mgp00004.html"><IMG SRC="mgp00004.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 4"></A>
+<A HREF="mgp00005.html"><IMG SRC="mgp00005.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 5"></A>
+<A HREF="mgp00006.html"><IMG SRC="mgp00006.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 6"></A>
+<A HREF="mgp00007.html"><IMG SRC="mgp00007.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 7"></A>
+<A HREF="mgp00008.html"><IMG SRC="mgp00008.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 8"></A>
+<A HREF="mgp00009.html"><IMG SRC="mgp00009.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 9"></A>
+<A HREF="mgp00010.html"><IMG SRC="mgp00010.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 10"></A>
+<A HREF="mgp00011.html"><IMG SRC="mgp00011.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 11"></A>
+<A HREF="mgp00012.html"><IMG SRC="mgp00012.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 12"></A>
+<A HREF="mgp00013.html"><IMG SRC="mgp00013.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 13"></A>
+<A HREF="mgp00014.html"><IMG SRC="mgp00014.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 14"></A>
+<A HREF="mgp00015.html"><IMG SRC="mgp00015.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 15"></A>
+<A HREF="mgp00016.html"><IMG SRC="mgp00016.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 16"></A>
+<A HREF="mgp00017.html"><IMG SRC="mgp00017.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 17"></A>
+<A HREF="mgp00018.html"><IMG SRC="mgp00018.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 18"></A>
+<A HREF="mgp00019.html"><IMG SRC="mgp00019.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 19"></A>
+<A HREF="mgp00020.html"><IMG SRC="mgp00020.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 20"></A>
+<A HREF="mgp00021.html"><IMG SRC="mgp00021.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 21"></A>
+<A HREF="mgp00022.html"><IMG SRC="mgp00022.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 22"></A>
+<A HREF="mgp00023.html"><IMG SRC="mgp00023.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 23"></A>
+<A HREF="mgp00024.html"><IMG SRC="mgp00024.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 24"></A>
+<A HREF="mgp00025.html"><IMG SRC="mgp00025.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 25"></A>
+<A HREF="mgp00026.html"><IMG SRC="mgp00026.idx.png" WIDTH=256 HEIGHT=192 ALT="Page 26"></A>
+<HR>
+Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+<BR>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00001.html b/papers/cuug2007/mgp00001.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00001.txt>[text page]</A> [&lt;&lt;start] [&lt;prev] <A HREF=mgp00002.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 1: Faster Packets<BR>
+<HR>
+<IMG SRC="mgp00001.png" WIDTH=1024 HEIGHT=768 ALT="Page 1"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00001.idx.png b/papers/cuug2007/mgp00001.idx.png
diff --git a/papers/cuug2007/mgp00001.png b/papers/cuug2007/mgp00001.png
diff --git a/papers/cuug2007/mgp00001.txt b/papers/cuug2007/mgp00001.txt
@@ -0,0 +1,10 @@
+
+
+
+
+Faster Packets
+10GE, faster pf and other fun stuff
+
+David Gwynne <[email protected]>
+Henning Brauer <[email protected]>
+
diff --git a/papers/cuug2007/mgp00002.html b/papers/cuug2007/mgp00002.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00002.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00001.html>[&lt;prev]</A> <A HREF=mgp00003.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 2: why 10Gb on OpenBSD?<BR>
+<HR>
+<IMG SRC="mgp00002.png" WIDTH=1024 HEIGHT=768 ALT="Page 2"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00002.idx.png b/papers/cuug2007/mgp00002.idx.png
diff --git a/papers/cuug2007/mgp00002.png b/papers/cuug2007/mgp00002.png
diff --git a/papers/cuug2007/mgp00002.txt b/papers/cuug2007/mgp00002.txt
@@ -0,0 +1,12 @@
+
+why 10Gb on OpenBSD?
+
+
+OpenBSD is popular as a router/firewall platform
+
+Networks and infrastructure are getting faster
+
+Hardware is now available
+
+People are going to want this
+
diff --git a/papers/cuug2007/mgp00003.html b/papers/cuug2007/mgp00003.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00003.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00002.html>[&lt;prev]</A> <A HREF=mgp00004.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 3: Hardware<BR>
+<HR>
+<IMG SRC="mgp00003.png" WIDTH=1024 HEIGHT=768 ALT="Page 3"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00003.idx.png b/papers/cuug2007/mgp00003.idx.png
diff --git a/papers/cuug2007/mgp00003.png b/papers/cuug2007/mgp00003.png
diff --git a/papers/cuug2007/mgp00003.txt b/papers/cuug2007/mgp00003.txt
@@ -0,0 +1,9 @@
+
+Hardware
+
+
+at a basic level its like 1Gb, but all the numbers are growing, eg the interface speed (duh), the size of the packets, the size of the descriptor rings, the number of descriptor rings, the price
+
+computers already struggle at 1Gb speeds to keep up with interrupts
+
+
diff --git a/papers/cuug2007/mgp00004.html b/papers/cuug2007/mgp00004.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00004.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00003.html>[&lt;prev]</A> <A HREF=mgp00005.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 4: Hardware<BR>
+<HR>
+<IMG SRC="mgp00004.png" WIDTH=1024 HEIGHT=768 ALT="Page 4"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00004.idx.png b/papers/cuug2007/mgp00004.idx.png
diff --git a/papers/cuug2007/mgp00004.png b/papers/cuug2007/mgp00004.png
diff --git a/papers/cuug2007/mgp00004.txt b/papers/cuug2007/mgp00004.txt
@@ -0,0 +1,8 @@
+
+Hardware
+
+
+10Gb is the same, but worse
+vendors are starting to throw more features (and bugs) in to make it easier on computers, eg, TOE, RSS/TSO, checksumming, RDMA
+most offer virtualisation "features", (apparently all of which are first to market and unique).
+
diff --git a/papers/cuug2007/mgp00005.html b/papers/cuug2007/mgp00005.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00005.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00004.html>[&lt;prev]</A> <A HREF=mgp00006.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 5: Vendors<BR>
+<HR>
+<IMG SRC="mgp00005.png" WIDTH=1024 HEIGHT=768 ALT="Page 5"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00005.idx.png b/papers/cuug2007/mgp00005.idx.png
diff --git a/papers/cuug2007/mgp00005.png b/papers/cuug2007/mgp00005.png
diff --git a/papers/cuug2007/mgp00005.txt b/papers/cuug2007/mgp00005.txt
@@ -0,0 +1,9 @@
+
+Vendors
+
+
+We know of eight
+S2IO/Neterion, Intel, Myrinet, Chelsio, NetXen, Tehuti, NetEffect, Sun
+
+We have cards from six, soon to be seven. Sun is the only hw we wont have.
+
diff --git a/papers/cuug2007/mgp00006.html b/papers/cuug2007/mgp00006.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00006.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00005.html>[&lt;prev]</A> <A HREF=mgp00007.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 6: Vendors<BR>
+<HR>
+<IMG SRC="mgp00006.png" WIDTH=1024 HEIGHT=768 ALT="Page 6"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00006.idx.png b/papers/cuug2007/mgp00006.idx.png
diff --git a/papers/cuug2007/mgp00006.png b/papers/cuug2007/mgp00006.png
diff --git a/papers/cuug2007/mgp00006.txt b/papers/cuug2007/mgp00006.txt
@@ -0,0 +1,14 @@
+
+Vendors
+
+
+Neterion are great, supplied hw/docs/drivers
+
+Tehuti are great, supplied hw/docs/engineers
+
+Intel are ok, supplied hw and a driver
+
+sun are yet to give up anything
+
+the others are supplying hw and lunix drivers.
+
diff --git a/papers/cuug2007/mgp00007.html b/papers/cuug2007/mgp00007.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00007.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00006.html>[&lt;prev]</A> <A HREF=mgp00008.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 7: 10Gb on OpenBSD<BR>
+<HR>
+<IMG SRC="mgp00007.png" WIDTH=1024 HEIGHT=768 ALT="Page 7"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00007.idx.png b/papers/cuug2007/mgp00007.idx.png
diff --git a/papers/cuug2007/mgp00007.png b/papers/cuug2007/mgp00007.png
diff --git a/papers/cuug2007/mgp00007.txt b/papers/cuug2007/mgp00007.txt
@@ -0,0 +1,10 @@
+
+10Gb on OpenBSD
+
+
+We have drivers for 3: xge (neterion), ixgb (intel), tht (tehuti)
+
+xge/ixgb are i386/amd64 only cos writing decent drivers is apparently too hard for some people.
+
+nx (netxen), che (chelsio), and myricom (myx) are in progress
+
diff --git a/papers/cuug2007/mgp00008.html b/papers/cuug2007/mgp00008.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00008.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00007.html>[&lt;prev]</A> <A HREF=mgp00009.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 8: 10Gb on OpenBSD<BR>
+<HR>
+<IMG SRC="mgp00008.png" WIDTH=1024 HEIGHT=768 ALT="Page 8"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00008.idx.png b/papers/cuug2007/mgp00008.idx.png
diff --git a/papers/cuug2007/mgp00008.png b/papers/cuug2007/mgp00008.png
diff --git a/papers/cuug2007/mgp00008.txt b/papers/cuug2007/mgp00008.txt
@@ -0,0 +1,10 @@
+
+10Gb on OpenBSD
+
+
+we only support the ethernet features, we don't use the offloading
+
+other oses can do like 2.5Gbps using the same limited features
+
+my box can do half a gigabit with no tweaks using tht
+
diff --git a/papers/cuug2007/mgp00009.html b/papers/cuug2007/mgp00009.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00009.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00008.html>[&lt;prev]</A> <A HREF=mgp00010.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 9: why that slow?<BR>
+<HR>
+<IMG SRC="mgp00009.png" WIDTH=1024 HEIGHT=768 ALT="Page 9"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00009.idx.png b/papers/cuug2007/mgp00009.idx.png
diff --git a/papers/cuug2007/mgp00009.png b/papers/cuug2007/mgp00009.png
diff --git a/papers/cuug2007/mgp00009.txt b/papers/cuug2007/mgp00009.txt
@@ -0,0 +1,10 @@
+
+why that slow?
+
+
+i have a cheap computer just for writing drivers, not for routing
+
+most chips support parallel operations. we cant do that cos of kernel locking
+
+our network stack is not optimized for single fast tcp streams (but very safe)
+
diff --git a/papers/cuug2007/mgp00010.html b/papers/cuug2007/mgp00010.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00010.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00009.html>[&lt;prev]</A> <A HREF=mgp00011.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 10: why not use the extra features<BR>
+<HR>
+<IMG SRC="mgp00010.png" WIDTH=1024 HEIGHT=768 ALT="Page 10"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00010.idx.png b/papers/cuug2007/mgp00010.idx.png
diff --git a/papers/cuug2007/mgp00010.png b/papers/cuug2007/mgp00010.png
diff --git a/papers/cuug2007/mgp00010.txt b/papers/cuug2007/mgp00010.txt
@@ -0,0 +1,14 @@
+
+why not use the extra features
+
+
+we dont trust the vendors stacks
+
+big reason is we like our network stack
+
+offload engines bypass pf
+
+we would have to complicate our stack to bypass it, which hurts everyone else
+
+offload engines cant scale like a real machine can (mem limits on a chip, etc), so end up on the machine anyway
+
diff --git a/papers/cuug2007/mgp00011.html b/papers/cuug2007/mgp00011.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00011.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00010.html>[&lt;prev]</A> <A HREF=mgp00012.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 11: what we've done?<BR>
+<HR>
+<IMG SRC="mgp00011.png" WIDTH=1024 HEIGHT=768 ALT="Page 11"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00011.idx.png b/papers/cuug2007/mgp00011.idx.png
diff --git a/papers/cuug2007/mgp00011.png b/papers/cuug2007/mgp00011.png
diff --git a/papers/cuug2007/mgp00011.txt b/papers/cuug2007/mgp00011.txt
@@ -0,0 +1,12 @@
+
+what we've done?
+
+
+two groups doing the same thing for different reasons
+
+dlg, reyk, claudio are focused on 10Gb
+
+henning and mcbride on making things faster just cos they think faster is better than fast
+
+but now we talk
+
diff --git a/papers/cuug2007/mgp00012.html b/papers/cuug2007/mgp00012.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00012.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00011.html>[&lt;prev]</A> <A HREF=mgp00013.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 12: making pf faster<BR>
+<HR>
+<IMG SRC="mgp00012.png" WIDTH=1024 HEIGHT=768 ALT="Page 12"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00012.idx.png b/papers/cuug2007/mgp00012.idx.png
diff --git a/papers/cuug2007/mgp00012.png b/papers/cuug2007/mgp00012.png
diff --git a/papers/cuug2007/mgp00012.txt b/papers/cuug2007/mgp00012.txt
@@ -0,0 +1,12 @@
+
+making pf faster
+
+
+pf is not exactly slow, but we thought it can be faster :)
+
+packets travel in so-called mbufs through the system
+
+mbufs are pre-allocated, getting them is cheap
+putting them back was kinda slow... dlg/tedu fixed that :)
+dlg's 10GE test box went from 700 to 900 MBit/s, just because of that: not calling microuptime() every time we put an mbuf back
+
diff --git a/papers/cuug2007/mgp00013.html b/papers/cuug2007/mgp00013.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00013.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00012.html>[&lt;prev]</A> <A HREF=mgp00014.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 13: making pf faster<BR>
+<HR>
+<IMG SRC="mgp00013.png" WIDTH=1024 HEIGHT=768 ALT="Page 13"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00013.idx.png b/papers/cuug2007/mgp00013.idx.png
diff --git a/papers/cuug2007/mgp00013.png b/papers/cuug2007/mgp00013.png
diff --git a/papers/cuug2007/mgp00013.txt b/papers/cuug2007/mgp00013.txt
@@ -0,0 +1,14 @@
+
+making pf faster
+
+
+pf needs to keep track of some things it does to packets
+save ip header position for altq
+queue ID for altq
+tag/tagged (as in, the keywords)
+ID of the routing table to be used for lookups
+count how often route-to kicked in, prevent loops
+mark packets generated by pf (block return, ...)
+mark packets that were rdr'd to localhost
+fragment cache needs a marker too
+
diff --git a/papers/cuug2007/mgp00014.html b/papers/cuug2007/mgp00014.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00014.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00013.html>[&lt;prev]</A> <A HREF=mgp00015.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 14: making pf faster<BR>
+<HR>
+<IMG SRC="mgp00014.png" WIDTH=1024 HEIGHT=768 ALT="Page 14"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00014.idx.png b/papers/cuug2007/mgp00014.idx.png
diff --git a/papers/cuug2007/mgp00014.png b/papers/cuug2007/mgp00014.png
diff --git a/papers/cuug2007/mgp00014.txt b/papers/cuug2007/mgp00014.txt
@@ -0,0 +1,13 @@
+
+making pf faster
+
+
+pf used to use mbuf tags for that.
+mbuf tags can contain arbitary information, attached to the mbuf, in chains so you can have multiple
+first used by ipsec
+
+in the old days, pf used up to eight distinct tags
+
+about a year ago, I folded that into one that has all info in it
+that already helps performance when you use that stuff, but I didn't measure back then
+
diff --git a/papers/cuug2007/mgp00015.html b/papers/cuug2007/mgp00015.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00015.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00014.html>[&lt;prev]</A> <A HREF=mgp00016.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 15: making pf faster<BR>
+<HR>
+<IMG SRC="mgp00015.png" WIDTH=1024 HEIGHT=768 ALT="Page 15"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00015.idx.png b/papers/cuug2007/mgp00015.idx.png
diff --git a/papers/cuug2007/mgp00015.png b/papers/cuug2007/mgp00015.png
diff --git a/papers/cuug2007/mgp00015.txt b/papers/cuug2007/mgp00015.txt
@@ -0,0 +1,15 @@
+
+making pf faster
+
+
+mbuf tags used malloc'd memory
+
+that turned out to be a major performance hit
+
+so stop using mbuf tags, and just stuff what we need into the mbuf header directly.
+grows the mbuf pkthdr slightly, but we have room there
+no mallocs!
+
+little sucky soekris box was capable of 29 MBit/s, with that change, 58.
+just love 100% increases
+
diff --git a/papers/cuug2007/mgp00016.html b/papers/cuug2007/mgp00016.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00016.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00015.html>[&lt;prev]</A> <A HREF=mgp00017.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 16: making pf faster<BR>
+<HR>
+<IMG SRC="mgp00016.png" WIDTH=1024 HEIGHT=768 ALT="Page 16"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00016.idx.png b/papers/cuug2007/mgp00016.idx.png
diff --git a/papers/cuug2007/mgp00016.png b/papers/cuug2007/mgp00016.png
diff --git a/papers/cuug2007/mgp00016.txt b/papers/cuug2007/mgp00016.txt
@@ -0,0 +1,8 @@
+
+making pf faster
+
+
+while I was working on that, chris kuethe found the sucky sis in the soekris can do interrupt mitigation, and turned that on in the driver
+
+dlg did some profiling, because his 10G card was so slow, and spotted the kernel randomness pool stirring from the network stack was a major hit
+
diff --git a/papers/cuug2007/mgp00017.html b/papers/cuug2007/mgp00017.html
@@ -0,0 +1,9 @@
+<HTML>
+<HEAD><TITLE>MagicPoint presentation foils</TITLE></HEAD>
+<BODY>
+<A HREF="index.html">[index]</A> <A HREF=mgp00017.txt>[text page]</A> <A HREF=mgp00001.html>[&lt;&lt;start]</A>  <A HREF=mgp00016.html>[&lt;prev]</A> <A HREF=mgp00018.html>[next&gt;]</A> <A HREF=mgp00026.html>[last&gt;&gt;]</A>
+<BR>Page 17: making pf faster<BR>
+<HR>
+<IMG SRC="mgp00017.png" WIDTH=1024 HEIGHT=768 ALT="Page 17"><BR>
+<HR>Generated by <A HREF="http://www.mew.org/mgp/">MagicPoint</A>
+</BODY></HTML>
diff --git a/papers/cuug2007/mgp00017.idx.png b/papers/cuug2007/mgp00017.idx.png
diff --git a/papers/cuug2007/mgp00017.png b/papers/cuug2007/mgp00017.png