Sizing hardward


(Jean-Christian BEDIER) #1

Hello,

We are studying the possibility to include elasticsearch in our websites
for search purpose (autocomplete).

Actually for our tests, we have 5 index which size is arround 200mb each.

Our search query is done through these 5 indexes for autocomplete.

We plan to have 2000 requests/seconds during pic seconds.

Do you have advice or recommendations about hardware sizing?

Shall we prefer more small servers or few fat servers ?

Thanks for your help.

Regards,

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


(ddorian43) #2

Since data<ram, i guess this should be easy scaled with replicas.

Record a few searches, benchmark agains different machines.

On Monday, October 14, 2013 3:46:08 PM UTC+2, Jean-Christian BEDIER wrote:

Hello,

We are studying the possibility to include elasticsearch in our websites
for search purpose (autocomplete).

Actually for our tests, we have 5 index which size is arround 200mb each.

Our search query is done through these 5 indexes for autocomplete.

We plan to have 2000 requests/seconds during pic seconds.

Do you have advice or recommendations about hardware sizing?

Shall we prefer more small servers or few fat servers ?

Thanks for your help.

Regards,

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


(Otis Gospodnetić) #3

Hi,

You won't need a lot of RAM, but because of that 2000 QPS, you will want
multiple CPU cores. How many exactly depends on how fast your autocomplete
queries (need to) run.

Imagine your requirement is to return AC suggestions within 1 second.
Imagine, for ease of calculation, each AC query takes 100 ms.
That means 1 CPU core can handle 10 of them in 1 second.
And you need to handle 2000 in 1 second, not 10.

So from here you can see how if your latency goes down you'll need fewer
CPU cores.

Otis

ELASTICSEARCH Performance Monitoring - http://sematext.com/spm/index.html
Search Analytics - http://sematext.com/search-analytics/index.html

On Monday, October 14, 2013 9:46:08 AM UTC-4, Jean-Christian BEDIER wrote:

Hello,

We are studying the possibility to include elasticsearch in our websites
for search purpose (autocomplete).

Actually for our tests, we have 5 index which size is arround 200mb each.

Our search query is done through these 5 indexes for autocomplete.

We plan to have 2000 requests/seconds during pic seconds.

Do you have advice or recommendations about hardware sizing?

Shall we prefer more small servers or few fat servers ?

Thanks for your help.

Regards,

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


(system) #4