Tom Lane wrote:
Stephen Marshall [EMAIL PROTECTED] writes:
2. The histogram concept is a neat idea, but I think some reorganization
of the page information might make it unnecessary. Currently the FSM
pages are sorted by BlockNumber. This was particularly useful for
adding
Tom,
I'm happy to see your attentions turning back to the FSM. I like the
design, but I do have a few suggestions, particularly about how to
handle oversubscription of the FSM.
1. When the FSM is oversubscribed and one is trying to decide which
pages to keep, remember that page info is
Stephen Marshall [EMAIL PROTECTED] writes:
1. When the FSM is oversubscribed and one is trying to decide which
pages to keep, remember that page info is stored in groups of
CHUNK_SIZE pages, where CHUNK_SIZE is current 32.
Right, oversubscription would actually need to be measured in chunks
Robert Treat [EMAIL PROTECTED] writes:
Now that indexes are getting some reporting, my understanding is an
index would report fewer pages overall than it's associated table, but
those pages would be completely empty. However, given that they don't
reported non-empty pages, the percentage of
On Thu, 2003-02-27 at 11:00, Tom Lane wrote:
Robert Treat [EMAIL PROTECTED] writes:
Now that indexes are getting some reporting, my understanding is an
index would report fewer pages overall than it's associated table, but
those pages would be completely empty. However, given that they
Robert Treat [EMAIL PROTECTED] writes:
I think I was thinking that a given table will always report more pages
than an index on that table, since tables can report 50% empty pages
while indexes only report 100% empty pages. This would cause tables to
generally be favored over indexes, even
I wrote:
Stephen Marshall [EMAIL PROTECTED] writes:
If we sort the page info by available space, we could then use binary
search to find space thresholds when we are handling oversubscription.
The list-of-chunks storage layout provides only limited traction for
searching anyway, and none
Stephen Marshall [EMAIL PROTECTED] writes:
If I understand the concept correctly, the histogram will only be
calculated when MultiRecordFreeSpace is called AND the FSM is
oversubscribed. However, when it is called, we will need to calculate a
histogram for, and potentially trim data from,
I've been thinking about improving the algorithm that the free space map
(FSM) uses to decide what to store when it's not got enough shared
memory to keep track of everything. The present design uses a dynamically
adjusted threshold for each relation, throwing away pages whose free
space amount
PS: Another idea I'm toying with is to dump out the FSM contents at
postmaster shutdown and reload them at restart, so that the FSM doesn't
have to start from ground zero on every restart cycle. But that's
independent of the management algorithm...
Correct me if I'm wrong, but the FSM is
Matthew T. O'Connor [EMAIL PROTECTED] writes:
Correct me if I'm wrong, but the FSM is only populated by vacuum, so there
is no FSM information for any given table / database until it's vacuumed, in
a long running production enviornment this may not be that important, but it
could result in a
11 matches
Mail list logo