cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
377
Views
0
Helpful
2
Replies

If "connection failure" is not a bug + Multi-Vlan SrvrFrm!

shukla1975
Level 1
Level 1

ACE-gurus & Gilles ,

First Query: We are facing "connection failure" counter increment, what all could be the reasons for this failures and what can be done to minimize it, if its not a bug, running 3.0 A2(1.0) ?

Second Query:I have a single farm with multiple vlans in it with URL hashing, we see that the load distribution on the caches in unequal, to say, 1 vlan with 1 server gets 500Meg of traffic and second vlan with 2 caches gets 400Meg of traffic which is almost equally divided between the 2 caches, its seems suspicious to the customer that the vlan is impacting load distribution, though we know its not possible to impact.

We do heavily suspect the kind of traffic entering the ACE like videos from various known sites goiing to 1 cache, on this with URl hash, does ace hash the whole URL, like www.abc.com/cat/video/ or just the domain, abc.com ??

IMHO its the whole URL, but need some confirmation on it, and to optimize somewhat equal distribution to the caches of traffic bytes ( as the no. of connections are almost equal.) what can be tweaked in the URl hash algorithm to make things work.

Have very little time to come up with a workable solution.

Thanx in advance

Shukla.

2 Replies 2

If the domain name is not in the URL, then there is no way ACE can match

the domain name by using only the URL.

For example if some one types in "http://www.reuters.com/resources/abc.htm"

then following header will be generated.

GET /resources/abc.htm HTTP/1.1

Host: www.reuters.com

User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1.14) Gecko/20080404 Firefox/2.0.0.14

Accept: text/css,*/*;q=0.1

Accept-Language: en-us,en;q=0.5

Accept-Encoding: gzip,deflate

....

....

here "/resources/abc.htm" will be the requested url and "www.reuters.com" will be the Host.

Url hash only takes url into consideration. If you want to use host to calulate hash then you will need

"predictor hash header Host" command.

Connection failure normally occurs due to backend server connections. There were few bugs open according to which this counter incremenents due to invalid client requests, when persistence rebalance is configured and no L7 matching occurs and in case of TCP resets. I am not sure if these are still open for A2(1).

You can play with "predictor hash url begin-pattern < > end-pattern < >" to improve the distribution.

Syed

Hi Iftekhar,

As per URL hash predictor from ACE point of view it hashes the whole URL till it reaches a question mark "?" , this has been confirmed.

As per my testing, all youtube videos have this URL+host header value.

http://www.youtube.com/watch?v=1w7K7zSFgfw

http://www.youtube.com/watch?v=nQqmbuQuKVc&feature=dir

Thus this site is hashed till

" www.youtube.com/watch?" and the rest is not taken into account, so all youtube traffic enters a single cache, which is not optimal.

Also we cant change the ACE to look beyond the "?"... thats another roadblock.

in this case dest-ip based hash predictor is a value proposition or not ?

PS: the multiple Vlan creation for the same serverfarm has NO impact on LB, tested.

thanx

Shukla.

Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: