practical limit on the number of UIDs
Shachar Shemesh
shachar at shemesh.biz
Wed Jun 29 13:44:10 IDT 2011
On 29/06/11 12:01, Muli Ben-Yehuda wrote:
> On Wed, Jun 29, 2011 at 11:43:00AM +0300, Tom Goren wrote:
>
>> Sounds like a job better suited for a DB backed authentication
>> system such as LDAP.
>>
>> This is just off the top of my head, but for some reason I wouldn't
>> want /etc/passwd and its relatives 1 million lines long.
> I shared the same instinctive response and am now trying to find hard
> data (ok, anecdotal data) to support it.
>
Often, when engineers ask a question, they seem to forget that their
baseline of comparison, most often manual labor, is not perfect either.
The most obvious example is NLP (natural language processing), where
when evaluating the accuracy of an NLP engine one should always remember
that humans parsing (reading) text have, on average, an accuracy of only
97% (down to 50% if text received via email, for some reason :-).
When you reach so many users, the problems relating to questions such as
"how long does it take nss to parse /etc/passwd" start to be dwarfed by
the human cost of maintaining a megaline text file. If for that reason
alone, you will need to switch to a DB back end.
I will remind you that when we opened Hamakor, we tried to manage the
list of members in a hand written CSV. At less than 20 members, the file
was no longer well formed and machine parseable. Text files are simply
not up to that task, regardless of parsing costs. An admin mistake will
wreak such havoc that it is simply impossible to reach that stage
anyways, and the core question is, IMHO, moot.
Shachar
--
Shachar Shemesh
Lingnu Open Source Consulting Ltd.
http://www.lingnu.com
More information about the Linux-il
mailing list