Estimating overview size

Chris Martin c.martin at sheffield.ac.uk
Sat Mar 11 20:38:03 UTC 2000


>>>>> On Thu, 9 Mar 2000 17:34:33 -0500 (EST), Bill Davidsen
 <davidsen at tmr.com> said:

> On 3 Mar 2000, Chris Martin wrote:
> In our development system 48% of a partition of 1164532 1k-blocks
> contains overview data for 1387329 articles.
> 
> The spool size is 3774828 blocks (cycbuffs) plus 55276 (tradspool) --
> a total of 3830104 1k-blocks.
> 
> Assuming that the ratio between spool size and overview space used
> holds, this means that the ratio of overview space used to spool size
> is 1:7 -- to allow space for expireover to work, allow 50% more
> overview space giving a ratio of 2:9.

Bill> I confess I usually lump history and overview in one f/s and
Bill> give it around 20% of the spool size to avoid problems. One
Bill> thing you have to watch is inodes, in 2.3 the tradindex takes at
Bill> least 1 inode per group, while bufindex seems to take less. Your
Bill> 2:9 sounds familiar, I like it.

Though as

>>>>> On Fri, 03 Mar 2000 12:34:15 -0800, Paul Theodoropoulos
  <paul at atgi.net> said:

Paul> here's my current reader box's salient info:

Paul> alt.binaries*   rawCNFS                 26G about 7-10hrs retention
Paul> alt.* + foreign heirarchies rawCNFS     26G about 25 days retention
Paul> * (all else) rawCNFS, mirrored  disks   26G about 45 days retention
Paul> local and regional in tradspool	       2G variable retention
Paul> rawdisk buffindexed overview            10G currently at 62% utilization

and

>>>>> On Fri, 03 Mar 2000 15:39:20 -0800, David Schwartz
  <davids at webmaster.com> said:

David> I have a 50Gb spool and only 1Gb for overview and history. But
David> then, I don't let any article stay over 14 days. What really
David> fills your overview and history are large numbers of small
David> articles. If you use CNFS and keep lots of small articles for
David> long periods of time, your overview/history will go through the
David> roof and performance will suffer a bit.

implied, it depends heavily on retention times -- though Paul's figures
give a 1:8 ratio...



More information about the inn-workers mailing list