Viewing Cache Performance
The Statistics applet serves as the CacheFlow's dashboard: pie charts,
histograms, and raw data provide insight into traffic volume, disk/memory
usage, cache efficiency, and stored content. Statistics are cumulative,
with a single "Clear Stats" reset buttonwe prefer this to GUIs that
lose history on reboot.
Volume (below, left)
presents histograms that measure objects, bytes, clients, CPU, and
freshness during the last hour, day, and month. Efficiency (below,
right) uses a pie chart to show distribution of objects or bytes
served as from the cache, loaded from source, or non-cacheable.
Volume panel
Efficiency panel
Others pie charts break
down non-cacheable content and contrast RAM vs. disk hits, while a
third panel substantiates all charts with raw data (below, left).
One metric we haven't yet seen in other products is the histogram
showing Content distribution (below, right). Knowing the size
of previously-stored objects can be helpful when tuning max HTTP and
FTP object size.
Efficiency: Data tab
Contents: Distribution tab
The Statistics applet also includes a General page that summarizes the
current configuration and provides operational status: temperature, fan,
power, and disk. We did not spot a network health monitor that might show
adapter status or gateway / DNS reachability, but significant network
events can be seen using the Event Viewer (see Advanced Monitoring).
Like other products, the CacheFlow logs client requests,
but it does so with a few twists. The log file must be uploaded to
a primary or alternate FTP server on a scheduled basis: daily, hourly,
on demand, or based on percent of available disk used (below, left).
There is no provision to "tail" the log while still on the cache,
or to store multiple logs on the cache. CacheFlow uses NCSA Common
Log Format by default, with Squid Format as an option, but also parses
a well-defined syntax to yield Custom Formats (below, right).
Log entries indicate more than HIT or MISS: they also signal object
expiration, forced refresh, filter/bypass blocking, and how responses
are obtained (e.g., direct from server, via ICP parent or sibling). Customization
can be handy to reduce the space consumed and focus on what is most relevant
to you.