Verify my Squid understanding

--MimeMultipartBoundary
Content-Type: TEXT/PLAIN; charset=US-ASCII

Hi, everybody,

        I am going to write a simulator to study the performance of the
removal policy implemented in Squid 1.1.20. Please correct my misunderstandings
of Squid. Thanks!

The environment I am interested:
1) Only one squid is running
2) My squid starts with a clean cache
3) Simulated time period is less than one year, so I avoid any calls to
storePurgeOld
4) No memory is used as a cache between proxy and the outside world, ie
all SwapIn, SwapOut functions are re-written to bypass the memory

This is my understanding of the Squid LRU policy:
1) Objects bigger than 4MB (configurable) will not be cached
2) Squid tries to run the purge algorithm (storeMaintainSwapSpace) once
per second.
3) storeEntries are stored in a hash table. When storeMaintainSwapSpace is
called, it only examines one of the buckets and then purges all the
purgeable storeEntries as defined in storeCheckExpired.
4) Only objects requested via GET method are cached
5) Only objects whose URL does not contain "?" or "cgi-bin" are cached
6) If store_swap_size > store_swap_high, storeGetSwapSpace will scan (only
count those scanned and unlocked) at most 256 entries stored in the hash
table and then sort them according to lastref time and finally remove the 8
entries that are least recently referenced every second (called only by
storeMaintainSwapSpace).
7) storeMaintainSwapSpace randomly chooses buckets to examine. Therefore
the Squid hit rate is not perfectly constant, although it falls within an
acceptable bound.

Yee Man Chan

--MimeMultipartBoundary--
Received on Tue Jul 29 2003 - 13:15:50 MDT