Skip to content

Commit

Permalink
minor edits
Browse files Browse the repository at this point in the history 
Signed-off-by: Ceki Gulcu <ceki@qos.ch>
  • Loading branch information
@ceki
ceki committed Aug 25, 2021
1 parent 41171f9 commit 0a97ddc
Show file tree
Hide file tree
Showing 2 changed files with 77 additions and 61 deletions.
32 changes: 19 additions & 13 deletions logback-site/src/site/pages/css/common.css
View file
Expand Up @@ -142,20 +142,15 @@ table.bodyTable td {
padding-bottom: 0.5ex;
}

table[class="bodyTable alignRight"] td {
text-align: right;
table.bodyTable.rightAlign tr td {
text-align: right;
}

/* apply to tr elements of tables which are both bodytable and dark */
table[class="bodyTable dark"] tr {
table.bodyTable[class="dark"] tr {
background-color: #ddd;
}

/* apply to tr elements of tables which are both bodytable and dark */
table[class="bodyTable properties"] tr {
vertical-align: top;
}

/* table.bodyTable tr.a { background-color: #ddd; } */
/* table.bodyTable tr.b { background-color: #eee; } */
/* table.bodyTable tr.alt { background-color: #eee;} */
Expand All @@ -173,7 +168,7 @@ table.bodyTable tr td p:first-child {
background-color: #f0f0f0;
}
.striped tr:last-child td {
border-bottom: 1px solid #ddd;
border-bottom: 2px solid #ddd;
}

td.word {
Expand Down Expand Up @@ -236,13 +231,24 @@ td.word {
}

.code {
font-family: Courier, monospace;
color: #666;
font-family: ui-monospace, SFMono-Regular, "SF Mono", Menlo, Consolas, "Liberation Mono", monospace;
background-color: #EEE;
}

.pattern {
font-family: ui-monospace, SFMono-Regular, "SF Mono", Menlo, Consolas, "Liberation Mono", monospace;
background-color: #DDD;
}

code {
font-family: ui-monospace, SFMono-Regular, "SF Mono", Menlo, Consolas, "Liberation Mono", monospace;
background-color: #EEE;
}


.variable {
font-family: Courier, monospace;
font-style:italic;
font-family: Courier, monospace;
font-style:italic;
}


Expand Down
106 changes: 58 additions & 48 deletions logback-site/src/site/pages/performance.html
View file
Expand Up @@ -43,7 +43,10 @@ <h2>Benchmarking synchronous and asynchronous logging</h2>
second or a service duration of a microsecond per operation. If
the destination of the event is a database with a throughput of
200 operations per second, the performance gain can be
tremendous.</p>
tremendous. This very favorable situation occurs if the rate of
logging event arrival is <em>slower</em> than the rate of
service, that is the rate at which events are transmitted to
destination and taken off the circular buffer.</p>

<p>Asynchronous logging is not a panacea however. It still has
to obey they laws of the physical universe. In particular, in
Expand Down Expand Up @@ -72,7 +75,7 @@ <h2>Benchmarking synchronous and asynchronous logging</h2>
logging events.
</p>

<p>When the <code>AsyncAppender</code> is overlaoded, your
<p>When the <code>AsyncAppender</code> is overloaded, your
application will be paying the cost of moving events in and out
of <code>AsyncAppender</code>'s circular buffer in addition to
sending events to destination. Thus, when we talk about
Expand All @@ -83,10 +86,17 @@ <h2>Benchmarking synchronous and asynchronous logging</h2>
contending for AsyncAppender's circular buffer.
</p>

<p>Therefore, an asynchronous logging benchmark should yield
information about the overhead of asynchronous logging (when
overloaded) and its behavior in presence of many producer
threads.
<p>As mentioned earlier, if the rate of event arrival is slower
than the service rate, the application will only have to
hand-off logging events to a circular buffer, significantly
lowering the perceived cost of logging. At the same time, we
must also consider the worst-case scenario where the rate of
logging event arrival is higher than the service rate. Such a
scenario is not that uncommon as slow appenders
abound. Therefore, as a worst-case analysis, an asynchronous
logging benchmark should yield information about the overhead of
asynchronous logging (when overloaded) and its behavior in
presence of many producer threads.
</p>

<p>We have chosen to use <code>FileAppender</code> in
Expand All @@ -108,10 +118,10 @@ <h3>Comparing log4j, log4j2 and logback</h3>
<p>We have taken care to compare apples to apples. All tests
specify the same buffer size, i.e. 256KB, for all
<code>FileAppender</code> instances across frameworks and using
the same identical pattern, that is "%d %p [%t] %c - %m%n".
Lossy behavior (when the buffer is full) is turned off for
logback's <code>AsyncAppender</code>. It is turned off by
default for log4j 1.2 as well as log4j2.</p>
the same identical pattern, that is <span class="pattern">"%d %p
[%t] %c - %m%n"</span>. Lossy behavior (when the buffer is
full) is turned off for logback's <code>AsyncAppender</code>. It
is turned off by default for log4j 1.2 as well as log4j2.</p>

<p>Regarding the circular buffer size for
<code>AsyncAppender</code> or log4j's LMAX ring buffer, we have
Expand All @@ -130,7 +140,7 @@ <h3>Comparing log4j, log4j2 and logback</h3>
<!--
easier cut and paste with emacs marco uusing forward-whitespace with </td><td> insertions
-->
<table class="bodyTable alignRight striped">
<table class="bodyTable rightAlign striped">
<tr>
<th>Threads</th>
<th>log4j synchronous</th>
Expand All @@ -150,7 +160,7 @@ <h3>Comparing log4j, log4j2 and logback</h3>
<td>844.67</td>
<td>2,139.83</td>
<td>1,760.30</td>
<td>"ops/ms"</td>
<td>ops/ms</td>
</tr>

<tr>
Expand All @@ -161,7 +171,7 @@ <h3>Comparing log4j, log4j2 and logback</h3>
<td>819.40</td>
<td>2,276.77</td>
<td>1,821.36</td>
<td>"ops/ms"</td>
<td>ops/ms</td>
</tr>
<tr>
<td>4</td>
Expand All @@ -171,49 +181,49 @@ <h3>Comparing log4j, log4j2 and logback</h3>
<td>770.27</td>
<td>1,836.99</td>
<td>1,799.39</td>
<td>"ops/ms"</td>
<td>ops/ms</td>
</tr>
<tr>
<td>8</td>
<td>633.13</td>
<td>726.21</td>
<td>1257.63</td>
<td>1,257.63</td>
<td>733.25</td>
<td>1787.62</td>
<td>1774.99</td>
<td>"ops/ms"</td>
<td>1,787.62</td>
<td>1,774.99</td>
<td>ops/ms</td>
</tr>
<tr>
<td>16</td>
<td>585.134</td>
<td>693.747</td>
<td>1211.313</td>
<td>722.345</td>
<td>1813.098</td>
<td>1815.107</td>
<td>"ops/ms"</td>
<td>585.13</td>
<td>693.74</td>
<td>1,211.31</td>
<td>722.34</td>
<td>1,813.09</td>
<td>1,815.10</td>
<td>ops/ms</td>
</tr>

<tr>
<td>32</td>
<td>643.851</td>
<td>657.082</td>
<td>1203.27</td>
<td>643.85</td>
<td>657.08</td>
<td>1,203.27</td>
<td>704.08</td>
<td>1782.814</td>
<td>1751.212</td>
<td>"ops/ms"</td>
<td>1,782.81</td>
<td>1,751.21</td>
<td>ops/ms</td>
</tr>

<tr>
<td>64</td>
<td>576.676</td>
<td>696.216</td>
<td>1236.377</td>
<td>726.153</td>
<td>1740.279</td>
<td>1644.819</td>
<td>"ops/ms"</td>
<td>576.67</td>
<td>696.21</td>
<td>1,236.37</td>
<td>726.15</td>
<td>1,740.27</td>
<td>1,644.81</td>
<td>ops/ms</td>
</tr>
</table>

Expand All @@ -239,19 +249,19 @@ <h3>Comparing log4j, log4j2 and logback</h3>
frameworks.
</p>

<p class="highlight">logback 1.3 performs about 3 times faster
than log4j and about 1.6 times faster than log4j2 in case of
synchronous logging. For asynchronous logging, logback 1.3
performs 2.5 faster than log4j and 2.3 times faster than
log4j2.</p>
<p class="highlight">logback version 1.3-alpha10 performs about
3 times faster than log4j and about 1.6 times faster than log4j2
in case of synchronous logging. For asynchronous logging,
logback 1.3 performs 2.5 faster than log4j and 2.3 times faster
than log4j2.</p>


<p>While these points may be somewhat subtle, the not so subtle
observation is that logback 1.3 performs about 3 times faster
than log4j and about 1.6 times faster than log4j2 in case of
synchronous logging. For asynchronous logging, logback 1.3
performs 2.5 faster than log4j and 2.3 times faster than
log4j2.</p>
observation is that logback version 1.3-alpha10 performs about 3
times faster than log4j and about 1.6 times faster than log4j2
in case of synchronous logging. For asynchronous logging,
logback 1.3 performs 2.5 faster than log4j and 2.3 times faster
than log4j2.</p>

<p>Note that with 2'200'000 events per second, and 209 bytes
written per event, logback's <code>FileAppender</code> is
Expand Down

0 comments on commit 0a97ddc

Please sign in to comment.