tag:blogger.com,1999:blog-6683631817010078964.post5295720453149246033..comments2023-05-02T15:02:34.970-07:00Comments on Gary Tully's blog: ActiveMQ - multiple kahaDB instances (mKahaDB) helping reduce journal disk usageGary Tullyhttp://www.blogger.com/profile/08884253042534204938noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-6683631817010078964.post-28991160991342217442015-03-04T07:19:11.552-08:002015-03-04T07:19:11.552-08:00Very intersting article.
I found that the use of ...Very intersting article.<br /><br />I found that the use of multiple KahaDB has many advantages. One of it is for PFC, if the PFC is enabled for reaching the limit of store (storeUsageHighWaterMark) and we start consuming messages, The PFC can be more reactive to detect the change with multiple KahaDB than the normal KahaDB.<br /><br />Does make sense ? What is your take on this ?Anonymoushttps://www.blogger.com/profile/15756326471055818728noreply@blogger.comtag:blogger.com,1999:blog-6683631817010078964.post-85773668484178487072012-08-14T04:54:28.715-07:002012-08-14T04:54:28.715-07:00@pieter, Yes, that makes sense.
It would be odd t...@pieter, Yes, that makes sense. <br />It would be odd that each ack was in a separate data file, but possible :-) .<br />Reducing the data file size reduces the probability of such a distribution.Gary Tullyhttps://www.blogger.com/profile/08884253042534204938noreply@blogger.comtag:blogger.com,1999:blog-6683631817010078964.post-57462064709312658162012-08-14T02:37:00.277-07:002012-08-14T02:37:00.277-07:00I am very interested in the point you are making w...I am very interested in the point you are making with acks locking files.<br />I would like to test my theory with you. Let's say, for the sake of argument we have datafiles of 32mb and each message takes up 1mb.<br />So 32 messages fit in one datafile.<br /><br />The active datafile is full, so 32 messages. Of which 31 are acknowlegded, we are waiting for the 32nd message to be acknowledged in order for the gc to do its job.<br />All acks for the 31 other messages are (each of them) in a different data file. So 31 files contain an ack for the active file.<br /><br />So that would mean that in my example 31 files are locked due to the acks + the active datafile = 32 datafiles.<br />At a rate of 32mb per file that would make a store of about 1024mb for only 1 unacked message.<br /><br />More generally put ((numberOfMessagesPerFile - 1) + 1 activeFile ) * dataFileSize = storage size<br />Again in my example ((32 -1) +1 ) * 32 = 1024mb.<br /><br />Does this make any sense? what is your take on this?<br /> <br />Anonymoushttps://www.blogger.com/profile/08940266451143542315noreply@blogger.com