Kim,

Thank you, i will play with that setting, please let me know if any other
tunings will help.

Ram

On Wed, Nov 28, 2018 at 8:04 AM Kim van der Riet <[email protected]>
wrote:

> The answer to your first question depends on what is more important to
> you - low latency or high throughput. Messages to be persisted will
> accumulate in a buffer page until it is full or until a timer is
> triggered, then it will be written to disk. It is not until this happens
> that the message will be acknowledged by the broker. If low latency is
> important, then having smaller but more numerous buffer pages will mean
> the messages will not wait for very long before being written to disk
> and acknowledged as received. However this occurs at the cost of some
> efficiency, which can affect throughput. If you have large volumes of
> messages and the throughput is more important, then using fewer but
> larger buffer pages will help you.
>
> Be aware, however, that the product of the size and number of pages is
> the total memory that will be consumed and held by the broker for
> buffering *per queue*. If you have a very large number of queues, then
> you must watch out that you don't over-size your write buffers or else
> you will run out of memory.
>
> While I cannot give you specific answers, as these depend on your
> performance priorities, I suggest some trial-and-error if you want to
> adjust these values.
>
> The Transaction Prepared List (TPL) is a special global queue for
> persisting transaction boundaries. As this info is usually small and
> relatively infrequent, the tpl-* settings apply to this queue only and
> the user has the option to use different values than the regular queues.
> If you don't use transactions, then this can be ignored. It is not a
> queue that can be written to directly, but the store creates its own
> data that is saved in this queue. Adjusting the tpl-* settings depends
> only on the frequency of transactions in the user's application or
> use-case.
>
> Hope that helps,
>
> Kim van der Riet
>
> On 11/27/18 4:44 PM, rammohan ganapavarapu wrote:
> > Kim,
> >
> > 1. My message size is around 80kb, so what would be suggested values for
> > the blow properties?
> >
> >
> > wcache-page-size
> > wcache-num-pages
> > tpl-wcache-num-pages
> > tpl-wcache-page-size
> >
> > right now i have all defaults, so i am trying to see if i can tune these
> > values for my messages size to avoid those AIO busy cases.  I have try to
> > define those properties/options in qpidd.conf file but when i run
> > qpid-config queues its not showing those values on my queues created by
> > client application, do i have to define those options when i create queue
> > instead of keep them in qpidd.conf?
> >
> > 2. What is difference b/w tpl-wcache-page-size and wcache-page-size
> >
> > Thanks,
> > Ram
> >
> > On Fri, Nov 16, 2018 at 9:26 AM Kim van der Riet <[email protected]>
> > wrote:
> >
> >> There is little documentation on linearstore. Certainly, the Apache docs
> >> don't contain much. I think this is an oversight, but it won't get fixed
> >> anytime soon.
> >>
> >> Kim
> >>
> >> On 11/16/18 12:11 PM, rammohan ganapavarapu wrote:
> >>> Any one point me to the doc where i can read internals about how
> >>> linearstore works and how qpid uses it?
> >>>
> >>> Thanks,
> >>> Ram
> >>>
> >>> On Mon, Nov 12, 2018 at 8:43 AM rammohan ganapavarapu <
> >>> [email protected]> wrote:
> >>>
> >>>> Kim,
> >>>>
> >>>> Thanks for clearing that up for me, does it support SAN storage
> blocks.
> >>>> Where can i read more about linearstore if i want to know the low
> level
> >>>> internals?
> >>>>
> >>>> Ram
> >>>>
> >>>> On Mon, Nov 12, 2018 at 8:32 AM Kim van der Riet <[email protected]
> >
> >>>> wrote:
> >>>>
> >>>>> The linearstore relies on using libaio for its async disk writes. The
> >>>>> O_DIRECT flag is used, and this requires a block of aligned memory to
> >>>>> serve as a memory buffer for disk write operations. To my knowledge,
> >>>>> this technique only works with local disks and controllers. NFS does
> >> not
> >>>>> allow for DMA memory writes to disk AFAIK, and for as long as I can
> >>>>> remember, has been a problem for the linearstore. With some work it
> >>>>> might be possible to make it work using another write technique
> though.
> >>>>> NFS has never been a "supported" medium for linearstore.
> >>>>>
> >>>>> On 11/9/18 4:28 PM, rammohan ganapavarapu wrote:
> >>>>>> But how does NFS will cause this issue, i am interested to see
> because
> >>>>> we
> >>>>>> are using NFS (V4 version) in some environments, so wanted to learn
> >>>>> tunings
> >>>>>> when we use NFS.
> >>>>>>
> >>>>>> Thanks,
> >>>>>> Ram
> >>>>>>
> >>>>>> On Fri, Nov 9, 2018 at 6:48 AM rammohan ganapavarapu <
> >>>>>> [email protected]> wrote:
> >>>>>>
> >>>>>>> Sorry, i thought it's NFS but it's actually SAN storage volume.
> >>>>>>>
> >>>>>>> Thanks,
> >>>>>>> Ram
> >>>>>>>
> >>>>>>> On Fri, Nov 9, 2018, 2:10 AM Gordon Sim <[email protected] wrote:
> >>>>>>>
> >>>>>>>> On 08/11/18 16:56, rammohan ganapavarapu wrote:
> >>>>>>>>> I was wrong about the NFS for qpid journal files, looks like they
> >>>>> are on
> >>>>>>>>> NFS, so does NFS cause this issue?
> >>>>>>>> Yes, I believe it does. What version of NFS are you using?
> >>>>>>>>
> >>>>>>>>
> >> ---------------------------------------------------------------------
> >>>>>>>> To unsubscribe, e-mail: [email protected]
> >>>>>>>> For additional commands, e-mail: [email protected]
> >>>>>>>>
> >>>>>>>>
> >>>>> ---------------------------------------------------------------------
> >>>>> To unsubscribe, e-mail: [email protected]
> >>>>> For additional commands, e-mail: [email protected]
> >>>>>
> >>>>>
> >> ---------------------------------------------------------------------
> >> To unsubscribe, e-mail: [email protected]
> >> For additional commands, e-mail: [email protected]
> >>
> >>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
>
>

Reply via email to