[Corpora-List] Limiting queries to online database

[William H Fletcher]

Thanks for your suggestions, Mark and Andrew.

I have now implemented limits of 1 query per second and 1000 queries total
per session per IP address.  That should meet the needs of most legitimate
researchers.  Since many people are loath to register I'll wait and see
whether I'll follow your example, Mark.  Requiring registration and log in
will certainly test users' motivation!

If any users on this list require a higher query limit I'll be glad to grant
it.  

BTW, the bots kept switching IP addresses; until I got my scripts tweaked
they ran over 12.5 M queries in the last couple of days, for over 21 M in
the last week!

Regards,
Bill 



-----Original Message-----
From: corpora-bounces at uib.no [mailto:corpora-bounces at uib.no] On Behalf Of
Hardie, Andrew
Sent: Monday, May 14, 2012 2:06 PM
To: corpora at uib.no
Subject: Re: [Corpora-List] Limiting queries to online database

Hi Bill,

My instinct would be to block the IP of any such nuisance clients
completely. 

The BNC is a very widely available dataset and if someone really does need
to run such a vast number of queries for a legitimate purpose, they should
be running them on a local copy and not leeching your bandwidth. Clearly
they have the tech savvy to do so if they are capable of programming this
robot. At the very least they should have had the courtesy either to
throttle the pace of queries right back, or to contact you to ask
permission. 

And that's giving them the benefit of the doubt and assuming they are
legitimate users to begin with; more likely than not it is some kind of spam
probe or, as Mark suggested, someone trying to avoid paying for the BNC. 

Although it won't deter malicious parties, you should consider using
robots.txt and a Crawl-delay directive so that genuine good-faith users know
what is an acceptable query rate. 

All of this, by the way, is one reason why with my own software, I recommend
people set up CQPweb servers with password-controlled access!

best

Andrew.

-----Original Message-----
From: corpora-bounces at uib.no [mailto:corpora-bounces at uib.no] On Behalf Of
William H Fletcher
Sent: 14 May 2012 18:27
To: corpora at uib.no
Subject: [Corpora-List] Limiting queries to online database

Hello,

My site http://phrasesinenglish.org/ provides query interfaces to databases
derived from the BNC. The past week performance has suffered from
extraordinarily high query traffic from a handful of IP addresses.  I am
seeking advice on what is a realistic limit to queries from one user and how
how to limit traffic from a single IP efficiently.

In the past my policy was to place no limits on number of queries or size of
datasets returned on the assumption that this generous approach facilitates
research.  Occasionally I have hit a sevrer with thousands of queries, but
at a maximum pace of 1-2 per second. Most users on the site submit at most a
few dozen queries per day.  On rare occasions I have seen short bursts of
say 20,000-60,000 queries from a single IP address.  

Last week over I found over 8 million queries for co-occurrences of
apparently random pairs of word forms (e.g. 'entertainer mussel') coming
from several IP addresses in Beijing.  Now, over the last day or so there
have been almost 6 million queries from one IP address in Seoul (110-120 per
second).  It's a valuable stress-test for my server, but I fear the
degradation of response times will drive away regular users.

1. What constitutes a reasonable number of queries per day to tolerate from
a single robot user, after which access would be denied or limited?  

2. How can I implement such access restrictions?  I am using the Nginx
server, MySQL / Sphinx and PHP on a Debian Linux platform.  I know how to
block an IP address completely, but have no good strategy for simply
limiting such traffic.

Many thanks in advance for any feedback you can give.

Regards,
Bill Fletcher


_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora

_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora


_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora