Differences

This shows you the differences between two versions of the page.

--- paolopisati:wraptest [2007/01/03 19:01] – PaoloPisati
+++ paolopisati:wraptest [2022/10/05 23:24] (current) – Move to personal namespace KubilayKocak
@@ Line 5: / Line 5: @@
   * Socket sizes: -s = -S = 57344.
   * Desktop: %%PentiumD%% 920 (2x2.8Ghz), 2Gb RAM - %%FreeBSD%%-HEAD/amd64.
-  * Wrap: AMD Geode (486 clone 233Mhz), 128MB RAM, 10/100 sis nics (DP83816 PCI) [http://www.pcengines.ch/wrap.htm details here].
+  * Wrap: AMD Geode (486 clone 233Mhz), 128MB RAM, 10/100 sis nics (DP83816 PCI) [[http://www.pcengines.ch/wrap.htm|details here]].
   * World and kernel for the wrap board were compiled with -O1 (except in the gcc optimization tests).
-  * A tarball containing test results, gnuplot files and plots is available [http://people.freebsd.org/~piso/wrap-exp/wrap-exps.tgz here].
+  * Kernel config file: [[http://people.freebsd.org/~piso/wrap-exp/WRAP-4.x|4.x]] and [[http://people.freebsd.org/~piso/wrap-exp/WRAP-7.x|7.x]].
+  * In the 7.x tests, ipfw was compiled as a module and it was not loaded during the simple forwarding tests.
+  * In the 4.x tests, i used 2 different kernels: one with IPFW & DIVERT compiled in (used only in the nat tests), and another without IPFW compiled in (used for all the other tests) - this way i avoided ipfw overhead in the forwarding tests.
+  * A tarball containing test results, gnuplot files and plots is available [[http://people.freebsd.org/~piso/wrap-exp/wrap-exps.tgz|here]].
+  * Below the images are my comments, feel free to add yours.
+  * Author: Paolo Pisati %%<%%piso@%%FreeBSD%%.org>
-http://people.freebsd.org/~piso/wrap-exp/wrap-exp.png
+{{http://people.freebsd.org/~piso/wrap-exp/wrap-exp.png}}
 **Legend:**
@@ Line 20: / Line 25: @@
   * poll-2x: device polling activated for sis0 and sis1
   * nod: net.isr.direct=0 (7.x default: net.isr.direct=1)
-  * natd: divert socket + natd
+  * natd: ipfw divert socket + natd
   * ipfwnat: ipfw nat
   * O0: -O0 (no optimization)
@@ Line 29: / Line 34: @@
 ----
-http://people.freebsd.org/~piso/wrap-exp/4vs7.png
+{{http://people.freebsd.org/~piso/wrap-exp/4vs7.png}}
   * In the 4.x-ffwd case, looks like the 100Mbit nics are the bottleneck.
@@ Line 35: / Line 40: @@
 ----
-http://people.freebsd.org/~piso/wrap-exp/4vs7-forwarding-polling.png
+{{http://people.freebsd.org/~piso/wrap-exp/4vs7-forwarding-polling.png}}
   * Dunno what's the cause for the drop in 4.x-fwd-poll2x.
   * In all the tests without polling enabled, the wrap board was unresponsive (livelock?!?!).
   * There's something definitely wrong with polling in 7.x (maybe since 5.x?).
+  * Device polling in 4.x gives a remarkable improvement.
 ----
-http://people.freebsd.org/~piso/wrap-exp/4vs7-fastforwarding-polling.png
+{{http://people.freebsd.org/~piso/wrap-exp/4vs7-fastforwarding-polling.png}}
   * All the 4.x plots are overlayed on top of each other.
-  * Turning on polling with fast forward enabled (contrary to what happened in the previous experiment), didn't give any performance boost: perhaps with Gb nics we would see an improvement?
+  * In 4.x, turning on polling with fast forward enabled (contrary to what happened in the previous experiment), didn't give any performance boost: perhaps with Gb nics we would see an improvement?
 ----
-http://people.freebsd.org/~piso/wrap-exp/4vs7-polling-complete.png
+{{http://people.freebsd.org/~piso/wrap-exp/4vs7-polling-complete.png}}
-  * Merge of the 2 previous experiments: quite messy, but useful to directly compare different combinations.
+  * Merge of the 2 previous experiments: **really messy**, but useful to directly compare different combinations.
 ----
-http://people.freebsd.org/~piso/wrap-exp/7-directvsnodirect.png
+{{http://people.freebsd.org/~piso/wrap-exp/7-directvsnodirect.png}}
   * As per rwatson request, i did this test (more from him? :).
@@ Line 62: / Line 68: @@
 ----
-http://people.freebsd.org/~piso/wrap-exp/4vs7-nat.png
+{{http://people.freebsd.org/~piso/wrap-exp/4vs7-nat.png}}
-  * In 4.x the firewall (ipfw) and fast forwarding are not compatible, while in 7.x, in the fast forward path, we traverse the pfil hooks so i turned it on in that case.
+  * In 4.x the firewall (ipfw) and fast forwarding are not compatible, while in 7.x (in the fast forward path) we traverse the pfil hooks so i turned it on in that case.
   * Turning on polling didn't improve the 4.x cases (while it improved simple forwarding - see above).
+  * In any case, 7.x ipfw nat looks faster than 4.x natd.
 ----
-http://people.freebsd.org/~piso/wrap-exp/7-opt-complete.png
+{{http://people.freebsd.org/~piso/wrap-exp/7-opt-complete.png}}
-  * Being the curious person i am, i wanted to see how much impact different gcc optimization levels can have on packets forwarding.
+  * Being the curious person i am, i wanted to see how much impact different gcc optimization levels can have on packets forwarding, and i was gladly surprised to see how the situation improved rising the optimization level - definitely worth a try imo, at least the -O1 level.
 ----
+**Conclusion:**
+  * 4.x (at least on slow/UP/this hw) is almost always the fastest solution.
+  * Turn on fast forwarding whenever possible.
+  * Device polling looks broken in HEAD (what about 5.x? 6.x?).
+  * 7.x ipfw nat is faster than 4.x natd.
+  * Rising compiler optimization level can significantly affect the performance.