Hi William- A lot goes into performance tuning a broker for various use cases. 5 minutes is not a long time to run a test either, I suggest running for a longer period of time for caches to be fully saturated.
First tip, I recommend measuring throughput in MB/s. Your two tests: 1024 x 1,700 = 1,740,800 MB/s 1,024,000 x 19 = 19,456,000 MB/s As you can see here, the 1MB message test is sending ~11x more data through the broker, even though the message-rate is lower. If you are using a transaction-per-message that is a pattern with very high network latency— which is what I suspect you are seeing. - Matt Pavlovich > On Feb 27, 2025, at 9:39 AM, William Crowell <wcrow...@perforce.com.INVALID> > wrote: > > Disclaimer before reading this: This is a tuning issue and not a performance > problem with ActiveMQ. > > > > Description of the issue: > > > > I believe I have a tuning problem with ActiveMQ 6.1.5 on Rocky Linux 9.5 > running on a virtual machine with 8 vCPUs and 32GB of RAM. The underlying > disk is attached SSD. > > > > I am using the ActiveMQ Classic Performance Module: > https://activemq.apache.org/components/classic/documentation/activemq-classic-performance-module-users-manual > > > > I am running the broker and the producer on separate virtual machine > instances. The producer is running on a 12 vCPU host. The broker and > producer VM instances are on the same VLAN. > > > > 2 scenarios: > > > > 1) Running a 5 minute test with 25 producer threads with 1K payloads on a set > of 50 queues I observed that I averaged about 1700 transactions/messages per > second using the producer provided with the plugin. > > > > 2) Running a 5 minute test with 25 producer threads with 1MB payloads on a > set of 50 queues I observed that I averaged about 19 transactions/messages > per second using the producer provided with the plugin. > > > > I really expected the throughput to be a lot more than what I am seeing. The > CPU utilization stays around 10-11% and there is plenty of free memory. I > did notice the disk saturation can get up to 80-90% during a test with write > activity. > > > > I am using the out of the box settings for the systemUsage segment in > activemq.xml which is 70% of the heap for memoryUsage, 100GB storeUsage, and > 50GB of tempUsage. > > > > I am going to try the same tests with the following settings: > > > > On the broker: > > > > Turning off journalDiskSyncStrategy (which I am not a real fan of doing): > > > > journalDiskSyncStrategy=”never” > > > > On the producer client running the Maven plugin: > > > > factory.optimizeAcknowledge=true > > factory.alwaysSessionAsync=false > > factory.prefetchPolicy.queuePrefetch=2000 > > factory.prefetchPolicy.queueBrowserPrefetch=1000 > > factory.useAsyncSend=true > > > > Is there anything I am forgetting to increase throughput? I feel like I am > leaving something silly out. > > > > Regards, > > > > William Crowell > > > > This e-mail may contain information that is privileged or confidential. If > you are not the intended recipient, please delete the e-mail and any > attachments and notify us immediately. >
