ES Hardware Profiles


(Michael Sick) #1

I'm looking for a baseline server recommendation for running ES in a
cluster. When I've reviewed other threads, there is little in the way of
specifics (though my search may not have been exhaustive). I understand
that there are many variables but I think a blessed baseline and some
specific guidelines would be helpful to those getting started and those
moving towards production.

A set of recommendations like:

http://webcache.googleusercontent.com/search?q=cache:UjPiR_xOhEwJ:hortonworks.com/best-practices-for-selecting-apache-hadoop-hardware/+&cd=4&hl=en&ct=clnk&gl=us
(sorry
the uncached version is password protected and I don't see a signup)

or

would be very helpful. Aside from the baseline, I'd like to better
understand:

  1. How does using ES as the primary data store impact the baseline?
  2. Baseline networking recommendations
  3. What's the max percent of data that should be on any one node for
    responsive fail over?
  4. Is there any role for a SSD drive on a node for fast swap? I remember
    Shay speaking of swapping Filter/Cache data to disk and it would seem
    reasonable for that use case. Didn't know if there were any current uses.

If we can get some consensus here on baseline and rules, I'd be more than
willing to write this up for posting on the ES site (or modifying /
updating an existing resource if needed).


(Gustavo Maia) #2

Hi Michael,

I have 2 servers on the Amazon, with the following configuration:

High-CPU Extra Large Instance

7 GB of memory
20 EC2 Compute Units (8 virtual cores with 2.5 EC2 Compute Units each)
1690 GB of instance storage
64-bit platform
I/O Performance: High
API name: c1.xlarge

Each instance has 4 hds.
But I'm finding the search a little slow when I'm indexing and
searching the same time.
I do not know if you have anything to do with the plugin to make copies in S3.
My ES is configured with two shards and one replica.
The seek time is 500ms up to a volume index of 1GB.

2011/12/2 Michael Sick michael.sick@serenesoftware.com:

I'm looking for a baseline server recommendation for running ES in a
cluster. When I've reviewed other threads, there is little in the way of
specifics (though my search may not have been exhaustive). I understand that
there are many variables but I think a blessed baseline and some specific
guidelines would be helpful to those getting started and those moving
towards production.

A set of recommendations like:
http://webcache.googleusercontent.com/search?q=cache:UjPiR_xOhEwJ:hortonworks.com/best-practices-for-selecting-apache-hadoop-hardware/+&cd=4&hl=en&ct=clnk&gl=us (sorry
the uncached version is password protected and I don't see a signup)

or
http://www.cloudera.com/blog/2010/03/clouderas-support-team-shares-some-basic-hardware-recommendations/

would be very helpful. Aside from the baseline, I'd like to better
understand:

How does using ES as the primary data store impact the baseline?
Baseline networking recommendations
What's the max percent of data that should be on any one node for responsive
fail over?
Is there any role for a SSD drive on a node for fast swap? I remember Shay
speaking of swapping Filter/Cache data to disk and it would seem reasonable
for that use case. Didn't know if there were any current uses.

If we can get some consensus here on baseline and rules, I'd be more than
willing to write this up for posting on the ES site (or modifying / updating
an existing resource if needed).

--
Gustavo Maia


(Michael Sick) #3

Hi Gustavo,

Thanks for the note - sorry for the lag. I did a demo with 8 servers of the
same configuration that you're using but was never able to tax them very
much (we were hitting limits on the HBase portion of the work well before
we were limited by ES). One thing I can say was that the performance was
pretty variable for writing to ES, often it screamed but it could lag on
the same test an hour later.

--Mike

On Fri, Dec 2, 2011 at 3:41 PM, Gustavo Maia gustavobbmaia@gmail.comwrote:

Hi Michael,

I have 2 servers on the Amazon, with the following configuration:

High-CPU Extra Large Instance

7 GB of memory
20 EC2 Compute Units (8 virtual cores with 2.5 EC2 Compute Units each)
1690 GB of instance storage
64-bit platform
I/O Performance: High
API name: c1.xlarge

Each instance has 4 hds.
But I'm finding the search a little slow when I'm indexing and
searching the same time.
I do not know if you have anything to do with the plugin to make copies in
S3.
My ES is configured with two shards and one replica.
The seek time is 500ms up to a volume index of 1GB.

2011/12/2 Michael Sick michael.sick@serenesoftware.com:

I'm looking for a baseline server recommendation for running ES in a
cluster. When I've reviewed other threads, there is little in the way of
specifics (though my search may not have been exhaustive). I understand
that
there are many variables but I think a blessed baseline and some specific
guidelines would be helpful to those getting started and those moving
towards production.

A set of recommendations like:

http://webcache.googleusercontent.com/search?q=cache:UjPiR_xOhEwJ:hortonworks.com/best-practices-for-selecting-apache-hadoop-hardware/+&cd=4&hl=en&ct=clnk&gl=us
(sorry

the uncached version is password protected and I don't see a signup)

or

http://www.cloudera.com/blog/2010/03/clouderas-support-team-shares-some-basic-hardware-recommendations/

would be very helpful. Aside from the baseline, I'd like to better
understand:

How does using ES as the primary data store impact the baseline?
Baseline networking recommendations
What's the max percent of data that should be on any one node for
responsive
fail over?
Is there any role for a SSD drive on a node for fast swap? I remember
Shay
speaking of swapping Filter/Cache data to disk and it would seem
reasonable
for that use case. Didn't know if there were any current uses.

If we can get some consensus here on baseline and rules, I'd be more than
willing to write this up for posting on the ES site (or modifying /
updating
an existing resource if needed).

--
Gustavo Maia


(system) #4