Liberator Performance

[chen]

Hi,

i have some question,
1. how many number of clients that server can handle so the latency is still small

I want to test performance (number of cilent and latency) assumed on this data :
Total Data 500 Object from 1 datasource.
Each client subscribed to 18 objects with total field in message is 211 field, so each object message contain around 10 - 20 field.
Per-user update rate of 10 updates/sec.

My Liberator configuration :
object-throttle-times 0.01
burst-min 0
burst-max 0.01



2. If I use Liberator FE, can I using Liberator in clustered server? and how to make it ?

3. How can I know Liberator EE pricing detail?


Thanks.

[garypalmer]

Hi Chen

Please find the answers to your questions below. Do not hesitate to ask if we can be of further assistance.

1. how many number of clients that server can handle so the latency is still small

A simple approximate answer to this question; it will be able to handle 10,000 clients comfortably, and probably double that (20,000) with some additional latency which may well be within your tolerance levels.

For a detailed discussion of Liberator benchmarking, please see our article, Caplin Liberator 4.4 Benchmarks, on freeliberator.com. This can be navigated to from the documentation page or click the following link; http://www.freeliberator.com/documentat ... s/HTMLdoc/

This covers various data rate/volume profiles, showing message latency and CPU utilisation for incresing numbers of users at those rates.

From you proposed usage profile of;

Total Data 500 Object from 1 datasource.
Each client subscribed to 18 objects with total field in message is 211 field, so each object message contain around 10 - 20 field.
Per-user update rate of 10 updates/sec.

My Liberator configuration :
object-throttle-times 0.01
burst-min 0
burst-max 0.01

It appears that our 'Medium update rates for large numbers of end users' benchmarks is the closest match to you proposed usage. Our 'Medium update rates for large numbers of end users' profile is;
1000 objects from 1 DataSource
Clients subscribe to 20 objects; each object has 5 fields (one with 15 characters and four with 5 characters - totalling 58 bytes per message)
All objects update twice per second
object-throttle-times not specified
burst min defaults to 0.1
burst max 0.1

The graph plotting number of clients against latency and CPU utilisation (on a HP ProLiant DL585 G2 machine) is included in the document. At a glance it indicates that the average message latency experienced when 10,000 users are connected is around 40ms, for 15,000 clients it is around 75ms, for 20,000 clients it is around 150ms and so on. These results will obviously vary on different hardware.

2. If I use Liberator FE, can I using Liberator in clustered server? and how to make it ?

Liberator FE does support clustering. Clustering is not configured on the Liberator; it is configured in the client.
A client will be configured to to failover from one Libertor to another if the first fails. The Liberator that the client fails over to will be connected to the same DataSource as the first Liberator and so will have the same data/object available to it. When a client fails over, the new Liberator may have to request data from the providing DataSource in order to continue servicing object updates that were being serviced by the first Liberator - so there may be minimal interruption in the objects' updates.

3. How can I know Liberator EE pricing detail?

For all purchasing and sales enquiries please contact out Sales Team by emailing sales@caplin.com

[chen]

ok Thanks, thats really help.


and can i ask some question again,

1. If I using apache in my web server,what kind of problems will appear in that server if that server handle connection of massive user (for example tens of thousands number of user)?

2. That user connection will be handled directly with liberator or apache?


Thanks.

[Martin Tyler]

Typically you would have both apache (or another webserver) and liberator. Apache to serve your website, eg html, javascript, images etc.. and Liberator dedicated to the streaming. So apache would only be doing what it normally does, serving up static files (or maybe php or whatever), Liberator would be handling the persistent connections.