[squid-users] office365 - brand new endpoint management question!!

[Jennifer Canterbury]

Did anyone see the changes to office365 IP and URLs management effective as
of 11/27/18? Microsoft appears to be be going to 3 different CDN networks
(Akamai, MarkMonitor, and ExactTarget) for office365 connectivity.

I have a few questions:
1) How many users do you think a single virtual squid server could
handle?
2) How many "acl bypass_parent" URLs can my cache squid proxy server
handle before browsing starts to slow?
3) We use a pac file to redirect office365 URLs to our squid box, does
anyone have any snazzy ways of getting-managing-updating the office365 URL
list in squid I would need to configure using the new MS web service?
(product idea!!)

Waiting with bated breath!

[joseph]

store id helper can minimize cdn links



-----
**************************
***** Crash to the future ****
**************************
--
Sent from: http://squid-web-proxy-cache.1019090.n4.nabble.com/Squid-Users-f1019091.html

[Amos Jeffries]

Post by Jennifer Canterbury
Did anyone see the changes to office365 IP and URLs management effective
as of 11/27/18?  Microsoft appears to be be going to 3 different CDN
networks (Akamai, MarkMonitor, and ExactTarget) for office365 connectivity.
1)  How many users do you think a single virtual squid server could
handle? 

Somewhere between 0 and infinity.

We do not measure HTTP agents in terms of "users" because one single
user can flood a network, or many tens of thousands per second be
handled easily. It depends entirely on what type of message any given
user is sending at any particular moment.

So HTTP is measured best in request-per-second (RPS). For most HTTP/1.1
traffic Squid can handle around 10-20K requests per second before
performance tuning needs to happen, after tuning twice that.

Post by Jennifer Canterbury
2)  How many "acl bypass_parent" URLs can my cache squid proxy server
handle before browsing starts to slow?

What is being done with these URLs? Simply having them configured does
not affect browsing. Testing the same few values repeatedly likewise is
not noticable for the better ACL types.

Using something like a regex ACL or testing random URLs from the set
without any pattern to tune for all result in quite low numbers before
traffic slows.


Amos