Scalaris down scaling and scalability

[Matthieu Ghilain]

Hi,

I would like to ask you two questions :

First how can I shutdown properly a Scalaris node so it does not leave
abruptly.

Secondly I have some Scalability problems (I am using revision 1939),
I am doing a test to check if the performance are growing with the
number of nodes (I make it vary from 4 -> 18). I have a machine
responsible of sending request to Scalaris nodes. This machine has one
connection to each Scalaris node in the ring. I have as much thread as
nodes in the ring so that requests are done in parallel.

I use small instance from Amazon for my testes (for the nodes
responsible of sending request and for the scalaris nodes), I put
below the specification from amazone

Small instance from amazon :
Small Instance – default*
1.7 GB memory
1 EC2 Compute Unit (1 virtual core with 1 EC2 Compute Unit)
160 GB instance storage
32-bit platform
I/O Performance: Moderate
API name: m1.small

I first write 20000 random keys and then I read every of those random
keys, so probably of conflict is really small and indeed I don't get
any failures in my test. I put below the graph showing the time taken
to perform this test according to the number of nodes I use.

[img]http://postimage.org/image/34vvj8obo/[/img]

This is not scaling for me, I can see that as the number of node
growths performance are not really improving.
I don't really get why.

I have tested with more powerful machines, performance are improved
but still as I get to 12 nodes performance do not improve much when
adding nodes.

I have also tested with two machine instead of one performing request,
performance increase slightly (something like one second less to
perform the same number of transactions).

By the way can Scalaris use several processors ? Because it does not
seem to use lot's of CPU when run on powerfull machines.

Do you have an idea on the origin of those results? Should I run
several instances of Scalaris on the same nodes? How much request in
parallel should I do to one scalaris node to get the best of it ?

Thank you for answering.

[Florian Schintke]

Hi,

your benchmark seems to be latency bound, as the Scalaris protocol,
with a separate client needs at least 6 message delays. If you put
more load on it, I guess, the larger instance can handle more load
before a client notices a latency increase due to system overload than
the smaller instance.

Florian

[Matthieu Ghilain]

> --
> You received this message because you are subscribed to the Google Groups "scalaris" group.
> To post to this group, send email to scal...@googlegroups.com.
> To unsubscribe from this group, send email to scalaris+u...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/scalaris?hl=en.

Florian

[Nico Kruber]

[Nico Kruber]

regarding scaling down:
At the moment, there is no API to gracefully shut down a Scalaris node. We
will add that once the graceful leave is stable enough (some bugs need fixing
first).

Nico