TREX Slave

[Former Member]

Hello All,

I have few question lingering in my head from quite a few days.... hopefully the TREX pundits herein can answer some...

1)How are multiple search requests handled in a slave search server.Assuming that there is a single slave server.

2)How are multiple search requests handled in a distrubuted architecture...Assuming that there are more than single slave server....

This would actually help us to deteremine if there is any benefit in hosting two slave instances on a single box or assigning a single slave to answer all the slave queries.

Regards

Aj

[Former Member · Accepted solution]

Hi ,

Check " Multiple Host " section in this link..

<a href="http://help.sap.com/saphelp_nw2004s/helpdata/en/a4/929d4206b70931e10000000a1550b0/frameset.htm">TREX</a>

You have more related info in service market place.

Regards,

Venkat.

[Reward points for useful answers]

[Former Member · Accepted solution]

Hi Ramana,

Thanks much for the reply!!

I have been through this help documnets my questions is more on the architectural perspective.

Regards

Ajey

[KarstenH · Accepted solution]

Hi Shiv,

- opening a second slave instance on the same hardware does not really provide for more fail safety => no reason to do that.

- if you have very large indexes and want to allocate more than 2 GB memory for the IndexServer process, you should rather start an additional IndexServer in the same slave instance => again no reason for a second slave instance on the same HW

What other reason to do so were you thinking of?

Regards, Karsten

[Former Member · Accepted solution]

Hi Karsten,

What do you actually mean by index server Process .... and how can i have multiple index serverprocesses started.

My idea was to have two slave instance so that they would behave indpendently and hence load balancing

Regards

aj

[Former Member · Accepted solution]

Hi Aj,

Each TREX installation on a host runs a set of services, i.e., speaking in technical terms: processes. One of these services/processes is the TREXIndexServer. So, what Karsten suggested is to rather start a second TREXIndexServer process than installing a complete second TREX on the same host. By the way, the latter would not be possible if both TREX instances should work together in the same landscape.

The second TREXIndexServer will help for distributing the indexes on that host amongst two processes such that the 2GB process size limit can be avoided, as described by Karsten.

In order to set up a second TREXIndexServer process on the same installation, first you have to make sure that you have another 2GB of main memory available. If this requirement is fulfilled, you stop your TREX instance, open the configuration file "TREXDaemon.ini" and in section [daemon] in the line programs=... add

, indexserver2

and restart TREX. Then, you might have to reorganize the index assignments. This can be quickly done using the TREXAdmin tool.

If you also want to increase search performance you should install a second TREX on a distinct host. Nevertheless, I have to tell you that there is no real load balancing. The queries from the application are sent round robin between all the different TREXIndexServers available in the landscape that are slaves (responsible) for the respective indexes addressed by these queries, i.e., the first query goes to the appropriate TREXIndexServer on host1, the second query goes to the TREXIndexServer on host2, etc. independent of the actual CPU load on the hosts.

I hope this helps. If you have further questions please refer to the various manuals for distributed TREX landscapes on the SAP Service Marketplace and also the pieces of information regarding this topic to be found on the SAP Help Portal.

Best regards,

Sascha.

[Former Member · Accepted solution]

Hi Sascha

What if you are interested in implementing "Note 937055 - TREX 6.1: Analysis of search requests using python script", and you only have one TREX server (with only one index server - no slaves) that performs well - would it then be advisable to run more TREXIndexServer processes on the same host? Can you run three TREXIndexServer processes on the same host or is the TREXIndexServer process only neccessary for the slave indexes? (Thus, we only need to run two TREXIndexServer processes).

Kind regards,

Martin

[Former Member · Accepted solution]

Hi Martin,

Theoretically speaking, you could run as many TREXIndexServer processes as you want as long as you provide 2GB of physical main memory per process! I think at this point we would get quite quickly to the limit. Furthermore, you should also think of the CPU power you would need. So, in a practical environment, I have not seen more than two TREXIndexServer processes started for the same TREX installation on one host.

Regarding your other questions, please note that if you only have one TREX installation it is not recommended at all to run the script from SAP Note 937055. You would run it on the master, which is not a good idea as the script will cause services to stop and start again. (I think the script even does not allow you to run it on a master installation.) This could harm your system, e.g., think of a sudden stop during an indexation process. In your described case with a well performing TREX, it is not necessary to start a second TREXIndexServer process as you do not seem to currently run into the problem of the 2GB process size limit. I guess at this point might be a misunderstanding because I think I can see your intention why you would want to start the second process. However, I have to tell you that you must not use this second TREXIndexServer process as a slave for an index whose master TREXIndexServer process runs on the same host. Sorry if this has not become clear from my former reply.

If you want to run the script from the above mentioned note in order to do some search analysis or statistics I would recommend to install a second TREX on another machine and use this second TREX as a slave. It would even be better to also install another TREX on another additional machine such that you have the master - slave -slave configuration that we recommend and let the master only do the indexing job but no searches.

Please allow me to add the remark that with SAP NetWeaver 2004s TREX 7.0, the performance analysis has become much easier and more convenient. It can all be controlled from the Python TREXAdmin tool. In general, technically the TREX 7.0 can be used in scenarios where you are currently using the TREX 6.1, however, this is dependent on the validation status from the respective application version used. When considering a change, however, keep in mind that you would have to do a new installation and reindex or migrate all your indexes. (Eventhough the indexes can be migrated, in most cases I suggest the actually easier way of reindexing.)

Best regards,

Sascha.

[Former Member · Accepted solution]

Hi Sascha

Thank you very much for a thorough answer! If you like, I don't mind creating a new thread and awarding you points through that.

BTW: I would like to subscribe to your contributions, so if you don't mind, then please go to Forums > Your Control Panel and activate the parameter called "allow other users to watch me".

Kind regards,

Martin

[Former Member · Accepted solution]

Hi Sascha,

Thanks a lot....I mean you have been more than helpfull.

As Sogaard rightly said your contributions will take SDN to greater heights.

Is there any documentation which describes this(TREX Index Processes ...please give pointers..

Regards

Aj

Message was edited by:

SAP

Message was edited by:

SAP

[former_member194626 · Accepted solution]

Hi Sascha,

My customer wants to setup a Master TREX server to do indexing and a slave server to do all searching. I have just installed TREX (AIX) on on server Standalone

With Global File Mgr and on other server TREX instance (mounted the /sapmnt) directory on this one so it readthe profile from first one. Using TREX 7.0 (NW2004s SR2 release)

Here is the setup

Server - SAPTXSD1 Instance - 02 SID TSD

Server - SAPTXTD1 Instance - 02 SID TSD (Mounted /sapmnt from SAPTXSD1 on SAPTXTD1, so read the Profile file of SAPTXSD1).

But when I open TREXAdmin Page I see only the individual servers on each system. How do I make such that one TREX server does indexing and other searching. Did I do anything wrong in the installation?