Re: [leafnode-list] Practical limits of leafnode 1.9.x

[Matthias Andree <matthias.andree@xxxxxx>]

Mark schrieb am 2003-10-13:

> I have read some conflicting info on the practical limits of leafnode.
> I have leafnode setup on a fairly new and fast pc and HD, and I am
> retrieving and archiving approx 50 newsgroups of which 3/4 are high
> volume (no binaries)
> 
> What I am interested in knowing is the typical maximum amount of users
> for leafnode in this type of environment. 

I don't know a rule of thumb.

> I have heard leafnode scales badly, but what exactly does this mean. Is
> leafnode limited to x amount of newsgroups vs x amount of users etc. Can
> leafnode scale up to 100 users in my given situation??

You'll have to find out - it depends: how many users are logged in to
leafnode at the same time, how much memory does the machine have, what
operating system does it run, how do the users' news readers behave,
what file system are you using and so on.

Leafnode is notorious for loading the whole group list into memory in a
session, and that might be one of the bigger concerns. I'd think serving
the articles itself is not that big an issue.

The limiting factor for "100 users" is memory. I do not intend to fix
this in leafnode-1.9, I may fix leafnode-2.0 though.

The limiting factor for "high-volume newsgroups" is the file system
efficiency. Directories with many entries are slow on traditional file
systems.

Some tuning advice for the information you're giving: leafnode stores
all articles for a group in a single directory, one file per article. If
these groups have many files, use a file system that uses directory
hashes or trees.  Candidates are ReiserFS, JFS, XFS (Linux for instance)
and FFS with DIRHASH option (recent FreeBSD versions support this). With
FFS, make sure you don't run out of inodes (check with "df -i").

I am not up to date with respect to HTREE support in ext2fs and ext3fs
on Linux, this might also work, but it might not yet be mature. Ask the
ext2 or ext3 maintainers for status information (and if you ask them,
please share the replies).

One other thing: leafnode-1.9's algorithm to retrieve articles by
Message-ID depends on the exact length of the file name. DO NOT
RECOMPILE LEAFNODE WITH DIFFERENT --with-spooldir SETTING IF YOU VALUE
YOUR SPOOL! Instead, when moving data to a different partition, leave
leafnode running as is and make /var/spool/news a symlink pointing to
the new location. Exception: if the path name length remains unchanged -
even artificially, f. i. /local/news/xxx/, this would work.

-- 
leafnode-list@xxxxxxxxxxxxxxxxxxxxxxxxxxxx -- mailing list for leafnode
To unsubscribe, send mail with "unsubscribe" in the subject to the list