How to run a specmail workload?

[elnino]

Hi,

I am running the specmail workload, but I encountered some problem.
I configure the mailserver as following steps:
1. Install the dovecot
2. run the makeme_mailserver.sh which is post in ExampleVM scripts to mount the disk and create the accounts.
3. Edit the /opt/SPECimap/IMAP_config.rc and set the CLIENTS variable to "mailserver:1200", execute the command :
     nohup ./run_load_gen.sh &
     ./run_init.sh
4. In the Client, execute the command
   java -jar clientmgr.jar -p 1094 -log
   java -jar clientmgr.jar -p 1088 -log
   java -jar specvirt -l



However, ERRORs occur (in the console which run "java -jar clientmgr.jar -p 1088 -log"):

2015-02-05 10:24:28:229 Warming up.
2015-02-05 10:24:38:791 Expanding MIME message body cache to 3073...
2015-02-05 10:32:46:183 Warning: .doMyLongCommand received an SocketTimeoutException exception
java.net.SocketTimeoutException: Read timed out
2015-02-05 10:32:46:183 ERROR: : Read timed out
2015-02-05 10:32:46:183 test413:Reconnecting user due to error condition during SEARCH_ALL_SUBJECT
2015-02-05 10:33:13:717 Warning: .doMyLongCommand received an SocketTimeoutException exception
java.net.SocketTimeoutException: Read timed out
2015-02-05 10:33:13:718 ERROR: : Read timed out
...

Do I miss any steps to run it? Please help me.
Thanks.

[ChrisFloyd]

Since it appears the workload has entered the warmup phase then receives some of these errors, this could potentially be a performance issue.  The SPECimap workload imparts a significant IO load on the SUT disk subsystem, and requires a SAN capable of handling a moderately heavy IO load. 

A equally likely scenario is that the dovecot mail server has not "warmed up" the mail indexes sufficiently.  For a new dovecot instance and a newly populated mailstore- it is often required to run for ~1 hour before the dovecot indexes are fully optimized.  During this time you can typically ignore mail errors.  Assuming this is what you are seeing, if the errors subside after 40 mins or so, you can stop the run, and backup (e.g., tar/gzip) the mailstore. This backup can then be unzipped to the mail directory prior to each run to ensure a properly warmed-up and consistent mailstore state.

-Chris

[elnino]

Hi,

How can I execute the "warmed up" period independently? And now I alter the value of WORKLOAD_LOAD_LEVEL(500) to 10 in SPECvirt/Control.config file, it seems to work well... the log is :

 2015-02-05 14:53:44:827 Adding clients...
-> 2015-02-05 14:53:44:827 Adding host[0]: client:1200
-> 2015-02-05 14:53:44:829 Initializing clients (this may take a few minutes)...
-> 2015-02-05 14:53:45:465 Waiting for SPECvirt start signal ...
-> 2015-02-05 14:53:45:465 sleeping for 1 seconds.
-> 2015-02-05 14:53:46:465 Starting clients...
-> 2015-02-05 14:53:46:467 Ramping up threads...
-> 2015-02-05 14:59:46:470 Warming up...
-> 2015-02-05 15:19:46:470 Clearing results...
-> 2015-02-05 15:19:46:473 Recording...
-> Attempted number of users = 10
-> 2015-02-05 15:19:46:483 Clearing results.
-> 2015-02-05 15:19:56:512,35,35,0,0,2413,0,2315
-> 2015-02-05 15:20:06:495,67,67,0,0,2490,0,2315
-> 2015-02-05 15:20:16:491,80,80,0,0,2534,0,2315
-> 2015-02-05 15:20:26:491,115,115,0,0,4520,0,2315
...

Thanks

[ChrisFloyd]

I've been most successful by significantly increasing the warmup time in Control.config (for example, WARMUP_SECONDS=3600).  If by the end of the warmup phase the errors have subsided, the mailstore indexes are probably warmed up, and the mailstore can be backed up to be used for future runs (without the need for such a long warmup in future runs).