-----Original Message-----
From: Derek Maxson [mailto:techmail(_at_)frontporch(_dot_)com]
Sent: Monday, April 02, 2001 3:42 PM
To: Yang, Lily L
Cc: ietf-openproxy(_at_)imc(_dot_)org
Subject: RE: Request for Comment: Elapsed-Time as Event in OPES/ICAP
2. However, there is no hard limit to number of users
per bucket. It just
works out that for 1000 users there are 25 per bucket.
I understand. This is just an example. The key is that only one group is
redirected at any given 15-second interval.
3. Are you suggesting that the proxylet would be
employed to evaluate every
request and measure the time, then create a call-out when the
time is right?
Yes. that is one way to do it, given that the proxylet has to be very simple
and efficient.
But I think there might be a better way to do this if we get enough
arithmetic support from IRML -- for example, I imagine that a rule like
this:
if (current-system-time-in-seconds mod 40 == Hash(IP))
fire-action-Proxylet1;
// basically it uses the current system timer (measured in seconds) to map
to a particular //user group (there are 40 such groups), and if the incoming
IP happens to be in that group,
// we can fire proxylet1 to do further checking.
Inside Proxylet1: do whatever further constraint checking -- you may even
afford to check on individual IP state (one counter per user), or you can
choose to use just one counter for the whole group -- then determine whether
or not redirection service is fired by insert new-tag-from-Proxylet1.
Second rule would be
if (new-tag-from-Proxylet1) fire-ICAP-service;
Does it make sense?
The second approach is more efficent (since it cuts down the # of requests
going through proxylet), but it requires more arithmetic support -- that
makes me wonder, is regular expression enough for rule matching?
Lily
Derek
-----Original Message-----
From: Yang, Lily L [mailto:lily(_dot_)l(_dot_)yang(_at_)intel(_dot_)com]
Sent: Monday, April 02, 2001 3:36 PM
To: 'Derek Maxson, Front Porch'
Cc: ietf-openproxy(_at_)imc(_dot_)org
Subject: RE: Request for Comment: Elapsed-Time as Event in OPES/ICAP
Derek --
Let me make sure that I understand what you meant here --
By splitting the users into 40 groups (bucket), we treat
two IPs falling
into one bucket as the same, so effectively it it like the
problem of
handling 40 unique users. Is it right?
So what you want to achieve is -- For every 15 seconds,
only IPs falling
into one group would be redirected for up to 25 connections?
Is it correct?
If so, then an unconditional rule is applied to fire up a local
proxylet and
all the above logic is wrapped inside that proxylet. Only
when redirection
(or whatever service in your mind) is determined necessary, the
remote call
out service is fired. In that case a new header is added in the
proxylet to
trigger the remote callout service in the second rule.
Is this what you had in mind?
Lily
-----Original Message-----
From: Derek Maxson, Front Porch
[mailto:techmail(_at_)frontporch(_dot_)com]
Sent: Friday, March 30, 2001 11:12 AM
To: ietf-openproxy(_at_)imc(_dot_)org
Subject: RE: Request for Comment: Elapsed-Time as Event
in OPES/ICAP
All:
Perhaps using groups for timing. A hash could be usedto
aggregate into
interval groups.
Take 1000 users to be redirected every 10 minutes.
Split the users into
40 groups by IP hash that will be redirected 25 at a time
every 15 seconds.
This would likely decrease the load of remembering time
stamps for all
1000 users (in this example) and enable much larger numbers
to be handled.
The proxy could keep track of each of the 40 groups as
buckets and load
balance by dropping IP's into the bucket(s) with the fewest
active sessions.
Thoughts?
Derek
-----Original Message-----
From: owner-ietf-openproxy(_at_)mail(_dot_)imc(_dot_)org
[mailto:owner-ietf-openproxy(_at_)mail(_dot_)imc(_dot_)org]On Behalf Of
Derek Maxson,
Front Porch
Sent: Monday, March 26, 2001 8:34 AM
To: ietf-openproxy(_at_)imc(_dot_)org
Subject: Request for Comment: Elapsed-Time as Event in OPES/ICAP
All:
A function that could prove helpful is to measure the
elapsed time between
ICAP callouts to ensure a minimum interval.
This function can allow the opportunity for individual
users to be
redirected via ICAP at certain intervals that can be
set in a rule.
Is anyone already working on something similar?
Comments?
Derek Maxson