[rabbitmq-discuss] Throughput and Latency.

[Michael Klishin]

On oct 3, 2013, at 6:54 a.m., k.madnani84 <k.madnani84 at gmail.com> wrote:

> That actually doesnt gives me a benchmark because the test is being executed
> in one of the fastest machine and throughput is very high.I was looking some
> considerable values.

Unfortunately, it's not possible to tell what "good throughput" or "good latency" without
knowing anything about

 * Median message size
 * Hardware used by nodes
 * Peak network capacity
 * Whether messages are published as persistent (and thus, have to be stored on disk)
 * How many queues are used in the workload
 * What client library is used
 * What exchange type is used (some are significantly less efficient algorithmically)
 * What concurrency rate is used with the load tool. What kind of hardware it uses.
 * TCP stack settings
 * What other software may be running on the same hardware

and so on.

On reasonably recent, medium tier hardware (say, last 3 years), it's known that a single
Rabbit node can demonstrate throughput in the 25-35K msg/s range for transient
messages. You can increase overall throughput by adding more nodes, more queues
and connecting directly to nodes that host master for a particular queue (to avoid
additional intra-cluster traffic routing).

Ultimately, the good number is the one that works well for your problem
and has some headroom for load peaks.

MK



-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 495 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20131003/83f146db/attachment.pgp>