Keep-Alive Information

This section provides information about the server’s HTTP-level keep-alive system.

---

Note –

The name keep alive should not be confused with TCP keep-alives. Also, note that the name keep-alive was changed to PersistentConnections in HTTP 1.1, but Web Server continues to refer to these connections as keep-alive connections. Most modern browsers request a web page from the server through persistent connections with the web server. The connection is kept alive even after processing a request, so that it will be easier to process a similar request.

---

The following example shows the keep-alive statistics displayed by perfdump:

KeepAliveInfo:
--------------------
KeepAliveCount        198/200
KeepAliveHits         0
KeepAliveFlushes      0
KeepAliveRefusals     56844280
KeepAliveTimeouts     365589
KeepAliveTimeout      10 seconds

The following table shows the keep-alive statistics displayed in the Admin Console:

Table 2–3 Keep-Alive Statistics

Number of Connections Processed	0
Total Number of Connections Added	198
Maximum Connection Size	200
Number of Connections Flushed	0
Number of Connections Refused	56844280
Number of Idle Connections Closed	365589
Connection Timeout	10

Both HTTP 1.0 and HTTP 1.1 support the ability to send multiple requests across a single HTTP session. A web server can receive hundreds of new HTTP requests per second. If every request is allowed to keep the connection open indefinitely, the server can become overloaded with connections. On UNIX and Linux systems, this can lead to a file table overflow very easily.

To resolve this problem, the server maintains a counter for the maximum number of waiting keep-alive connections. A waiting keep-alive connection has fully completed processing the previous request, and is now waiting for a new request to arrive on the same connection. If the server has more than the maximum waiting connections open when a new connection waits for a keep-alive request, the server closes the oldest connection. This algorithm keeps an upper bound on the number of open waiting keep-alive connections that the server can maintain.

Web Server does not always honor a keep-alive request from a client. The following conditions cause the server to close a connection, even if the client has requested a keep-alive connection:

• The keep alive timeout is set to 0.

• The keep alive maximum connections count is exceeded.

• Dynamic content, such as a CGI, does not have an HTTP content-length header set. This applies only to HTTP 1.0 requests. If the request is HTTP 1.1, the server honors keep-alive requests even if the content-length is not set. The server can use chunked encoding for these requests if the client can handle them (indicated by the request header transfer-encoding: chunked).

• The request is not HTTP GET or HEAD.

• The request was determined to be bad. For example, if the client sends only headers with no content.

The keep-alive subsystem in Web Server is designed to be massively scalable. The out-of-the-box configuration can be less than optimal if the workload is non-persistent (that is, HTTP 1.0 without the KeepAlive header), or for a lightly loaded system that’s primarily servicing keep-alive connections.

Keep-Alive Count

This section in perfdump has two numbers:

• Number of connections in keep-alive mode, also known as the total number of connections added

• Maximum number of connections allowed in keep-alive mode simultaneously, also known as the maximum connection size

Tuning

You can tune the maximum number of connections that the server allows to wait at one time before closing the oldest connection in the Admin Console by editing the Maximum Connections field on the configuration's Performance tab ⇒ HTTP tab, under Keep Alive Settings. The default is based on the number of available file descriptors in the system. In the command-line interface, use the max-connections property in the wadm set-keep-alive-prop command.

---

Note –

The number of connections specified by the maximum connections setting is divided equally among the keep-alive threads. If the maximum connections setting is not equally divisible by the keep-alive threads setting, the server might allow slightly more than the maximum number of simultaneous keep-alive connections.

---

Keep-Alive Hits

The keep-alive hits, or the number of connections processed, is the number of times a request was successfully received from a connection that was kept alive.

This setting is not tunable.

Keep-Alive Flushes

The number of times the server had to close a connection because the total number of connections added exceeded the keep-alive maximum connections setting. The server does not close existing connections when the keep-alive count exceeds the maximum connection size. Instead, new keep-alive connections are refused and the number of connections refused count is incremented.

Keep-Alive Refusals

The number of times the server could not complete the connection to a keep-alive thread, possibly due to too many persistent connections (or when total number of connections added exceeds the keep-alive maximum connections setting). The suggested tuning is to increase the keep-alive maximum connections.

Keep-Alive Timeouts

The number of times the server closed idle keep-alive connections because client connections timed out without any activity. This statistic is useful to monitor; no specific tuning is advised.

Keep-Alive Timeout

The time, measured in seconds, before idle keep-alive connections are closed. Set this value in the Admin Console in the Timeout field on the configuration's Performance tab ⇒ HTTP tab, under Keep Alive Settings. The default is 30 seconds, meaning the connection times out if idle for more than 30 seconds. The maximum is 3600 seconds (60 minutes). In the command-line interface, use the timeout property in the wadm set-keep-alive-prop command.

Keep-Alive Poll Interval

The keep-alive poll interval specifies the interval in seconds at which the system polls keep-alive connections for further requests. The default is 0.001 second, the lowest value allowed. It is set to a low value to enhance performance at the cost of CPU usage.

To tune the poll interval, edit the Poll Interval field on the configuration's Performance tab ⇒ HTTP tab, under Keep Alive Settings. In the command-line interface, use the poll-interval property in the wadm set-keep-alive-prop command.

Keep-Alive Threads

You can configure the number of threads used in the keep-alive system in the Admin Console by editing the Threads field on the configuration's Performance tab ⇒ HTTP tab, under Keep Alive Settings. The default is set to the number of processors in the system. In the command-line interface, use the threads property in the wadm set-keep-alive-prop command.

Tuning for HTTP 1.0-Style Workload

Since HTTP 1.0 results in a large number of new incoming connections, the default acceptor threads of 1 per listen socket would be suboptimal. Increasing this to a higher number should improve performance for HTTP 1.0-style workloads. For instance, for a system with 2 CPUs, you might want to set it to 2. You might also want to reduce the keep-alive connections, for example, to 0.

HTTP 1.0-style workloads can have many connections established and terminated.

If users are experiencing connection timeouts from a browser to Web Server when the server is heavily loaded, you can increase the size of the HTTP listener backlog queue by setting the HTTP listener listen queue size to a larger value, such as 8192.

The HTTP listener listen queue specifies the maximum number of pending connections on a listen socket. Connections that time out on a listen socket whose backlog queue is full fail.

Tuning for HTTP 1.1-Style Workload

While tuning server-persistent connection handling, balancing throughput and latency is a challenge. The keep-alive poll interval and timeout control latency. Lowering the value of these settings is intended to lower latency on lightly loaded systems, for example, to reduce page load times. Increasing the values of these settings is intended to raise aggregate throughput on heavily loaded systems, for example, by increasing the number of requests per second the server can handle. However, if there's too much latency and too few clients, aggregate throughput suffers as the server sits idle unnecessarily. As a result, the general keep-alive subsystem tuning rules at a particular load are as follows:

• If there's idle CPU time, decrease the poll interval.

• If there's no idle CPU time, increase the poll interval.

Also, chunked encoding could affect the performance for HTTP 1.1 workload. Tuning the response buffer size can positively affect the performance. A higher response buffer size in the configuration's Performance tab ⇒ HTTP tab would result in sending a Content-length: header, instead of chunking the response. To set the buffer size using the CLI, use the wadm set-http-prop command's output-buffer-size property.

You can also set the buffer size for a Service-class function in the obj.conf file, using the UseOutputStreamSize parameter. UseOutputStreamSize overrides the value set using the output-buffer-size property. If UseOutputStreamSize is not set, Web Server uses the output-buffer-size setting. If the output-buffer-size is not set, Web Server uses the output-buffer-size default value of 8192.

The following example shows using the CLI to increase the output buffer size, then deploying the configuration (used if UseOutputStreamSize is not specified in obj.conf):

./wadm set-http-prop --user=admin-user --password-file=admin-password-file 
--config=config-name output-buffer-size=16384
./wadm deploy-config --user=admin-user --password-file=admin-password-file
 --config=config-name

The following example shows setting the buffer size for the nsapi_test Service function:

<Object name="nsapitest">
 ObjectType fn="force-type" type="magnus-internal/nsapitest"
 Service method=(GET) type="magnus-internal/nsapitest" fn="nsapi_test"
 UseOutputStreamSize=12288
</Object>