We are running an ActiveMQ production environment, a Windows 2008 Server running on a VMWare ESX host, stored on an EMC SAN.
For all other purposes this system is quick and responsive, but the KahaDB fsync call/EnableJournalDiskSync is a major problem for performance. With the default on setting we get about 3MB/sec, if we turn it off we get 300MB or above - a 100x increase. This is most likely related to the fsync call not returning before the data has passed through the SSD cache on the SAN and stored on the RAID 6 disks. However, since we have an enterprise-grade SAN with SSD and battery backup, connected to a UPS we're probably less likely to lose data than if we were running on a normal server? If the data have reached the SAN SSD cache we should be pretty safe? We've already researched using multiple KahaDB stores and spliting the queues, but even that will most likely not give us the performance headroom we need for growing the next 2-3 years. Any input on wether or not we could turn EnableJournalDiskSyncs off would be greatly appreciated :) Thanks, Nikolai -- View this message in context: http://activemq.2283324.n4.nabble.com/EnableJournalDiskSyncs-slow-on-SAN-safe-to-disable-tp4681110.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
