Re: memory mapped store entries

From: Stewart Forster <slf@dont-contact.us>
Date: Tue, 25 Aug 1998 14:01:51 +1000

Hiya,

        One other thing to consider for mmap()ed storage index files will
be the disk load required to sync this data out.

        Currently squid just writes batches of storage updates at the end
of the store index file for each cache dir. This should mean one 8K write
per 8K/56(?) =~ 125 object updates.

        An mmap() implementation would presumably be jumping about all over
the object index changing stuff. Assuming 30 seconds between flushes we
could be talking about needing to rewrite the entire index file every 30
seconds for the worst case. For us that's 10M at its smallest. That's a
lot of extra disk IO.

        If you could ensure good locality WITHIN the updates to the mmap()ed
index then this won't be so much of an issue, but it definately is a
consideration for high end systems.

        Stew.

-- 
Stewart Forster (Snr. Development Engineer)
connect.com.au pty ltd, Level 9, 114 Albert Rd, Sth Melbourne, VIC 3205, Aust.
Email: slf@connect.com.au   Phone: +61 3 9251-3684   Fax: +61 3 9251-3666
Received on Tue Jul 29 2003 - 13:15:52 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:53 MST