Section D: News cache performance

About news cache tuning

A news cache does not accept or send news feeds. Therefore, a news cache administrator does not need to closely monitor disk usage, as is necessary for a news server. In addition, the administrator does not need to manage the news cache's interaction with news servers. After you configure your news caches, administration of the news caches is minimal.

When to use the NNTP access log

Use the NNTP access log if you want to view any of the following:

Errors during transactions (NNTP codes)
Number of groups and articles that a client accessed
Time, in seconds and milliseconds, between a client connection and disconnection
Number of bytes received from the client
Number of bytes sent to the client
Errors that NetCache generated
Errors that the server generated
Timestamp (in seconds and milliseconds)

Why the cache hit rate cannot be calculated for news caching

The cache hit rate is the number of user requests that NetCache resolves by using an object in the cache instead of fetching the object from the Web server. This definition is true for the cache hit rate for Web caching because a single Web request from a client remains a single request from the time that the client issues it to the time the client receives the requested data.

In contrast, one news request from the client can become multiple requests to the news server and, therefore, multiple responses from the news server. Therefore, the cache hit rate cannot be calculated for a news cache the same way as it can for a Web cache.

The following illustration compares how NetCache sends a Web request and a news request to the server.

NNTP bandwidth

NetCache calculates the percentage of bandwidth saved and displays it on the NNTP - General page under the Data tab of the Appliance Manager.

The fewer times that a news cache contacts a news server, the lower the bytes- from-server value.

NNTP statistics

The Appliance Manager displays the current status of NNTP request traffic and general statistics on cached and uncached articles served to clients. The statistics allow you to assess how well the news cache is functioning.

Consider the number of requests sent to the server on the NNTP - General page, and the request volume on the NNTP - Request Volume page under the Data tab.

Options that affect resolution time of news objects

The following configuration options on the on the NNTP - Caching Behavior page under the Setup tab in the Appliance Manager affect the response time for news requests, as well as the bandwidth savings:

Canceled Article Time-To-Live option
This is the number of seconds that you want to elapse before, upon user request, NetCache checks the host specified in the NNTP Server option to determine if an article in the cache was canceled. The default value is 3600 seconds.
NNTP Refresh Rate option
This is the number of seconds to elapse before NetCache checks the host specified in the Server IP Address and Port option for the arrival of new articles in a news group. The default value is 600 seconds (10 minutes).

The current status of NNTP request traffic and general statistics for cached and uncached articles returned to clients can be seen on the NNTP - General page under the Data tab in the Appliance Manager.

For details about the Canceled Article Time-To-Live and NNTP Refresh Rate options, see About the setup options.

Effect of the configured value on response time and bandwidth savings

For the Canceled Article Time-To-Live and NNTP Refresh Rate options, the higher the value you set, the less often NetCache contacts the news server. The less often NetCache has to check with the news server, the faster the response time and the better the bandwidth savings rate. However, some objects delivered to users from the cache might be stale.

Regardless of your settings of the Canceled News Article Time-To-Live and NNTP Refresh Rate options, NetCache verifies the consistency of an object whenever the period specified in either of these two options expires and the object is requested by a user.

Effect of adding an FQDN to the NNTP log

Adding a client's fully qualified domain name (FQDN) by customizing (adding c-dns to the NNTP access log header) can result in reduced performance because of the additional time needed to perform DNS lookups. Increased latency also occurs when the FQDN lookup is added to other logs.

The NNTP access log header can be checked for the presence of the c-dns lookup on the NNTP - Logging page under the Setup tab in the Appliance Manager.

Consideration if the news cache points to another news cache

You can configure one news cache to point to another news cache rather than to a news server. For example, you might want to have news caches at POPs point to a news cache at a data center to reduce the number of connections to the news server. If one news cache is pointing to another, the NNTP Refresh Rate time period is the sum of the NNTP Refresh Rate values that are set on the first news cache and the other news cache that it points to. The same is true for the Canceled Article Time-To-Live option.

Increasing the probability of multiple requests for the same article

The average size of a news article is less than 4k, the block size of a NetCache appliance. However, the alt.binaries.* hierarchy routinely carries articles that exceed 100k. The alt.binaries.* hierarchy tends to include the most popular newsgroups at most sites.

If your cache contains many large objects, adding disk space to a system already fully populated with disks can increase the probability that articles in the cache will be requested by more than one user.