Monitoring CMS Tracker construction and data quality using a grid/web service based on a visualization tool
G. Zito, M.S.Mennea, A. Regano, UNIVERSITY & INFN , Bari, Italy
The complexity of the CMS Tracker (more than 50 million channels to monitor) now in
construction in ten laboratories worldwide with hundreds of interested people , will
require new tools for monitoring both the hardware and the software. In our approach
we use both visualization tools and Grid services to make this monitoring possible.
The use of visualization enables us to represent in a single computer screen all
those million channels at once. The Grid will make it possible to get enough data
and computing power in order to check every channel and also to reach the experts
everywhere in the world allowing the early discovery of problems .
We report here on a first prototype developed using the Grid environment already
available now in CMS i.e. LCG2. This prototype consists on a Java client which
implements the GUI for Tracker Visualization and a few data servers connected to the
tracker construction database , to Grid catalogs of event datasets or directly to
test beam setups data acquisition . All the communication between client and servers
is done using data encoded in xml and standard Internet protocols.
We will report on the experience acquired developing this prototype and on possible
future developments in the framework of an interactive Grid and a virtual counting
room allowing complete detector control from everywhere in the world.
The complexity of the CMS Tracker (more than 50 million channels to monitor) now in construction in ten laboratories worldwide with hundreds of interested people , will require new tools for monitoring both the hardware and the software. In our approach we use both visualization tools and Grid services to make this monitoring possible.
The use of visualization enables us to represent in a single computer screen all those million channels at once. The Grid will instead make it possible to get enough data and computing power in order to check every channel.
Deploying this visualization tool as a Grid service means also that it can be used everywhere in the world enabling the experts to be in constant touch with the detector.
This allows the early discovery of problems in hardware and software.The tool developed
is used in this early stage to monitor the detector construction and the simulated data quality.
We report here on a first prototype developed using the Grid environment
already available now in CMS i.e. LCG2. This prototype consists on a client
which implements the GUI for Tracker Visualization . This part is implemented in Java and
can run on any computer connected to Internet . When this program needs data
to be displayed, it uses Grid and Web services to get them. These data
may come from a database interfaced to the Web (like the tracker construction
database) or from events in some dataset registered in the RLS.
The data are provided
from servers that access and analyze them using CMS software and
send the result to the visualization client. All the communication between client and servers is done using data encoded in xml and standard Internet protocols.
The client can display a generic tracker defined as a set of "modules"
organized in a hierarchy with rings , layers ,etc ...
The structure of the detector is read from a set of xml files available
from the same server that provides the event data .In this way the client can be used to monitor a series
of test prototypes in addition to the final complete detector.
We will report on the experience acquired developing this prototype and
present our opinion about some questions concerning CMS software in the field of interactive analysis and monitoring and the possible evolution of this project:
- Is a virtual counting room with complete detector control from everywhere in the world feasible?
- How difficult is to transform an "offline" application in a Web/grid service?
- How relevant are interactive Web and Grid services for CMS ?
- What is the best way to integrate CMS software with Web/Grid interactive
The system should consist of a highly portable lightweight client capable of
running on many different platforms connected to Internet.The term lightweight
implies that the client should be independent of the huge legacy libraries making up the main reconstruction and analysis code.
The server application should incorporate standard analysis tools and can be in direct contact with the event store or other databases. The connection
protocol between client and server should be standardised.
The usual way to represent Hep detectors and events is 3D with standard
projections and "layered projections" to provide both good overall pictures for
public relations and images useful for event analysis.In our case , the sheer
complexity of the tracker makes these standard representations not sufficient.
Successful monitoring requires representations where you can see the whole detector with each one of 17000 modules visible. For this reason we have developed
a new 2D representation.This is the main representation for detector monitoring
and is supplemented by the normal 3D/2D representation.
Why Web services
The use of web services is essential to ensure the interoperability of our visualisation application with different data sources.Some of these sources (the
event store) are on the Grid and their use requires Grid services which are
based on Web services. Other data sources are relational databases.
In both cases Web services let you establish a standard protocol for data
exchange between disparate data sources and a consumer.They serve as a layer
of abstraction isolating from the details of data representation, exchange and
query. For example the visualisation client handles in the same way tracker
data coming from the construction relational data base and from a grid dataset.
The architecture of the developed system can be seen in this picture.
The system is of the client-server type with a highly portable
lightweight client capable of running on many different platforms
connected to Internet. The server application works as a gateway
in direct contact with the construction database or the event
store where standard analysis program (ORCA - Cms reconstruction
project) write their data. The connection between client and
server uses standard internet protocols. To extract tracker data
from xml files we use the technique called xml data binding
as implemented by free source software Castor .
Visualization client implementation
The visualization client , which we named Tmon, is implemented in Java. It can execute on any computer
with Java Runtime 1.4 installed. This allows the use of our software from any
computer connected to Internet without the need to install special software.
This client communicates with data servers using standard Internet protocols.
All files exchanged contain data encoded in xml.
This client implements the user graphics interface and fulfills these main
- Present to the user all available data servers.
- Present for each data server, the available datasets.
- Once the user has selected a data server, request the tracker description
to the server.
- Interpret the data in the xml files that describe the tracker, configuring
its graphics interface for the specific tracker.
- Allow the user to request new events.
- Manage the events sent by the data servers (in xml format).
- Display the information sent in a 2D window representing either the
complete tracker or the single layer.
- Display the information sent in a 3D window. (This may be implemented later).
- Allow many different ways of displaying information.
- Allow the possibility to print the graphical representation on the screen.
- Allow the possibility to select any part of the visualization getting
information about the data.(This may be implemented later).
To extract tracker data from xml files we have used the technique called
xml data binding using the free source program Castor. This program uses
directly the schema definition file to create
a set of Java classes that can be used to read (unmarshall)
the tracker data from a xml file like this and
to have access to them in the visualisation client.
The client is available for download from Internet.
Data servers implementation
As data server we have used Tomcat + Axis.
Tomcat can be used as a normal Web server using http protocol. The client requests events or other data to the server by putting the
information about the request directly in the URL with the GET directive.
In some cases the POST directive is used and a file containing the information
in XML format is sent.The server answers by sending the data in the XML format.
To get events from Montecarlo federations we have used the so called REST approach to Web services. This consists in naming all resources provided with standard Web addresses(URI) and using the commands already available in http protocol(i.e. GET,PUT,DELETE,POST) to manage the resources. Tomcat implements all of them. We use instead Axis (working as a Tomcat servlet) to provide web services following the W3C definition with SOAP and WSDL. This approach is used
to serve the data in the construction database.
Up to now the data servers can provide data from three different
- - federations of Montecarlo events for the full detector.
- - real data from a prototype in a test beam(not yet implemented)
- - data from the construction data base
Serving data from the construction database
The construction database is based on a Oracle relational database.
This data base accepts any sql query embedded in a special xml command.
The result is a xml files containing the answer to our query.
To get the data we have implemented an Axis web service named
tracker.This service will accept requests for information in the
data base as normal URI and transform them in the xml/sql request to
the oracle database. The resulting xml file is transformed in the
standard cms tracker event format and sent to the visualization
client using SOAP. The web interface is described by a WSDL file.
The service is of the type "document" and has the following operations implemented:
- getTracker which returns the whole tracker geometry
definition in xml.
- getEvent which returns latest data in database
Number of dead strips in each module represented with a color code.
Serving data from Montecarlo federations
The data server connected to Montecarlo events is implemented with a
Tomcat servlet that basically answers the request
get next event in the tracker.
The event is taken from a local disk cache and returned in xml format.
The servlet will get the events by querying a number of grid computers.
The events may come from any number of computers. Each available
computer will run a simple Orca application that generates new Montecarlo
events and writes the tracker rechits in a disk area accessible also to the Tomcat servlet. The servlet checks from time to time if some new event is available
and , if this is true, it copies the event to the local cache.
Example of Montecarlo event(around 100000 rec hits overlayed view) represented by Tmon.Same event but separated mode view.
Detector and event data representation in xml
The XML files describing the various tracker layouts follow the tracker data model
developed by the same authors to implement a detailed tracker visualization
in the CMS offline software.
A generic tracker can be described as a hierarchy of modules.
Starting from the single module (a trapezoidal or rectangular box) we have
the following groupings : rings, layers, detector parts(i.e. endcap+z, endcap-z, barrel), subdetector (i.e. pixel detector,silicon inner detector, silicon outer detector), full tracker.
This structure is mapped by the following hierarchy of xml tags:
<module id="0" xcenter="-0.553762" ycenter=-"0.094102"
zcenter="-2.6308" type="0" />
<module id="1" xcenter="-0.560815" ycenter=-"0.0315087"
zcenter="-2.6278" type="0" />
<moduletype id="0" length="0.1151" width="0.0714" thickness="0.0003"
widthAtHalfLength="0.06465" nStrips="512" />
Here is the schema definition . This
is instead the description of the whole tracker.
N.B. We don't use directly the DDD : this is too complex for our purposes
and also incomplete. We use a
simpler tracker description and populate it with the data extracted (once for
all) from the DDD .
We have also defined the XML structure for any type of data coming from
the tracker : event data, monitoring data, data from construction.
Here is an example of event data containing reconstructed hits in the tracker.
Here a complete event.
<recHit globalX="-0.557684" globalY="-0.0389969" globalZ="-2.6375"
localX="-2.34123" localY="0" errorX="0.0191961" errorY="11.04" />
Xml data binding with Castor and web service implementation with Axis
The relationship between Axis and Castor can be seen in this
Castor uses this description of
the data model to generate automatically the java classes implementing
it (Castor beans).
The web service itself is described in the WSDL language in this file.
This file uses the XML-Schema file as
a definition of the way the Web service should transfer data to
The program WSDL2Java (provided with the Axis
distribution) uses the WSDL description of the Web service to
generate the Java code implementing it.
Serialization/Deserialization (converting data to/extracting data
from) xml files is done by Castor.
Although not present here, the event data model and the event transfer in the web service is implemented in the same way.
Note also that the only difference , at the level of data transfer protocol , between REST and SOAP services is that in SOAP services the events and tracker description are sent included in a SOAP envelope. The XML schema used is the same in both
The data servers are run on a PC Pentium IV.
We have tried the visualisation client in Java on many Linux and Windows PC with Java Runtime 1.4
installed . The time needed to fetch the description of the whole tracker
and unmarshal it is 10 seconds. This time is reasonable taking in account that
the xml file sent is more than 1 Mb large and the Java program has to validate
and load the data .Reading a single event or the data from the construction
database will take a time that depends on the data size but never exceeds
10 seconds. This time is obtained of course because both reading events
and data from database are done by servlets that work as gateway to the
real data provider caching the result of query.
The table below gives some examples of response time with events of different
sizes.(We don't do any optimization, compressing, etc or use any trick to decrea
se this time).
| # rechits in event||xml file size(MB)||
Lessons learned and conclusion
This was first of all a didactic exercise to get ready for the Grid.In fact
the next step is to trasform the Web service in a Grid service.
In this second phase we plan to try to use frameworks like MonALISA[6} and
Clarens already used to provide experimental Grid enabled data analysis
To implement our software we had to learn a lot of new technologies
mostly connected to Java and Xml and it can be of interest to know
what were our difficulties grasping these new technologies and also
to know our opinion on these.
We will report on this considering separately the various tasks that
we had to manage:
Anyhow all the software used has proven
to be reliable and with a good performance.But at the end you ask yourself if undertaking
this ordeal was really necessary. Our answer is yes! Xml is the future.
It is the future like Java was the future only a few years ago: perhaps
Java has been a failure but its main concept was so compelling that Microsoft
has reinvented it with .Net. To illustrate this point we would like to speak about another xml dialect discovered doing this work, SVG(Scalable Vector Graphics).
SVG is very interesting in our application because it can replace completely
the Java client! For this reason we are investigating its use.
How does it work? The servlet instead of sending the data will send a complete
image of the tracker with the data in the SVG format. This image can be
read by a Web browser (with an appropriate plugin) and has all the code to
allow the user to interact with the image including zooming,selection, etc...
Yes Xml is the future.
We had to convert an existing interactive graphics application from C++ and Qt to
Java : this was ,as you would imagine, very,very easy.
- We had to reimplement in Java the tracker object mode already implemented
in C++ : after we learned the trick of writing in XML our tracker object
model and using Castor to build the classes implementing it, it was really
a question of hours.Nothing to do with the months spent with C++.
- We had to write a servlet and deploy it with Tomcat : this proved to be
difficult if not very difficult. A servlet is not easy as an applet and if you
think so you are for some big surprise. Developing web applications in Java
is a lot more difficult that developing them for example in PHP.
A servlet container like Tomcat can work a lot better than PHP but only
in experts hands.
- We had to develop a Web service: this is really very difficult and we get stuck
on this for months. XML may look very simple but to use it in real life you
need to learn a plethora of xml languages and techniques: XML-Schema, SOAP,
WDSL, XSLT ,DOM,SAX,...
In addition to this the implementation of the Web service with Axis
proved to be rather tricky because we wanted to use Castor generated classes and the document style of web service.
Is monitoring a detector like CMS tracker feasible from Internet?
Yes. The results obtained are promising and indicate that an expert can have
a detailed update of the situation in the control in a few minutes.
The tracker map used as a convenient way to summarize monitoring data
about the tracker, is essential since it limits the amount of data that have
to travel between server and client.
But essential to the success of this schema is also the development by CMS of a standard Web and Grid interface to
its monitoring data. Developing our application we have found that the
access to construction data base was very easy because Oracle databases
have a standard Web interface. Instead the access to CMS event data
was very difficult because CMS has no Web interface defined and we had
to invent it from scratch by ,for example, defining a XML format of
CMS events.In our opininion , developing such standard Web interface, is
absolutely essential to CMS especially if it wants to fully exploit the
capabilities of the Grid. There are efforts going in this direction at least in the offline analysis environment: these are based on software like
the monitoring framework MonALISA and Clarens:it must be checked if they answer
to the requests made by Cms tracker online monitoring.
- Mennea,M.S; Regano,A; Zito,G. "CMS Tracker Visualisation" , CMS-NOTE-2004-009; Geneva : CERN,08Jun2004
- Events used by tmon
- http://www-106.ibm.com/developerworks/webservices/library/ws-castor/ : Create Web services using Apache Axis and Castor
- First draft of poster in pdf:
- poster in ppt format
- First draft of paper in pdf(in the Proceedings of IEEE NS Rome Oct.2004)
- Second draft of paper in pdf(Submitted to Transactions on Nuclear Science )