Has anyone run Accumulo on a single server straight from memory?
Probably using something like a Fusion  IO drive.  We are trying to use
it without using an SSD or any spinning discs.

 

Matthew Moore

Systems Engineer

SAIC, ISBU

Columbia, MD

410-312-2542

 

I have run a small cluster with HDFS writing only to a RAM disk.  Is that
the sort of thing you are interested in?

-Eric

On Tue, Sep 11, 2012 at 12:02 PM, Moore, Matthew J. <
[EMAIL PROTECTED]> wrote:

> Has anyone run Accumulo on a single server straight from memory?  Probably
> using something like a Fusion  IO drive.  We are trying to use it without
> using an SSD or any spinning discs.****
>
> ** **
>
> *Matthew Moore*****
>
> Systems Engineer****
>
> SAIC, ISBU****
>
> Columbia, MD****
>
> 410-312-2542****
>
> ** **
>

You could mount a RAM disk and point HDFS to it.

On Tue, Sep 11, 2012 at 9:02 AM, Moore, Matthew J. <[EMAIL PROTECTED]
> wrote:

> Has anyone run Accumulo on a single server straight from memory?  Probably
> using something like a Fusion  IO drive.  We are trying to use it without
> using an SSD or any spinning discs.****
>
> ** **
>
> *Matthew Moore*****
>
> Systems Engineer****
>
> SAIC, ISBU****
>
> Columbia, MD****
>
> 410-312-2542****
>
> ** **
>

Woops- slow innurnet and didn't notice Eric's response.

On Tue, Sep 11, 2012 at 9:30 AM, William Slacum <
[EMAIL PROTECTED]> wrote:

> You could mount a RAM disk and point HDFS to it.
>
>
> On Tue, Sep 11, 2012 at 9:02 AM, Moore, Matthew J. <
> [EMAIL PROTECTED]> wrote:
>
>> Has anyone run Accumulo on a single server straight from memory?
>> Probably using something like a Fusion  IO drive.  We are trying to use it
>> without using an SSD or any spinning discs.****
>>
>> ** **
>>
>> *Matthew Moore*****
>>
>> Systems Engineer****
>>
>> SAIC, ISBU****
>>
>> Columbia, MD****
>>
>> 410-312-2542****
>>
>> ** **
>>
>
>

Have you tried it where you're writing to straight block memory?  Not
using any file system or SATA controller.

 

Matt
Sent: Tuesday, September 11, 2012 12:19 PM
To: [EMAIL PROTECTED]
Subject: Re: Running Accumulo straight from Memory

 

I have run a small cluster with HDFS writing only to a RAM disk.  Is
that the sort of thing you are interested in?

 

-Eric

On Tue, Sep 11, 2012 at 12:02 PM, Moore, Matthew J.
<[EMAIL PROTECTED]> wrote:

Has anyone run Accumulo on a single server straight from memory?
Probably using something like a Fusion  IO drive.  We are trying to use
it without using an SSD or any spinning discs.

 

Matthew Moore

Systems Engineer

SAIC, ISBU

Columbia, MD

410-312-2542

 

 

Accumulo needs something that provides the FileSystem interface.  It also
needs to be distributed, replicated, and provide for a write-ahead log.
 HDFS on a RAM disk pretty much gets you that.

On Tue, Sep 11, 2012 at 12:55 PM, Moore, Matthew J. <
[EMAIL PROTECTED]> wrote:

> Have you tried it where you’re writing to straight block memory?  Not
> using any file system or SATA controller.****
>
> ** **
>
> Matt****
>
>
> *Sent:* Tuesday, September 11, 2012 12:19 PM
> *To:* [EMAIL PROTECTED]
> *Subject:* Re: Running Accumulo straight from Memory****
>
> ** **
>
> I have run a small cluster with HDFS writing only to a RAM disk.  Is that
> the sort of thing you are interested in?****
>
> ** **
>
> -Eric****
>
> On Tue, Sep 11, 2012 at 12:02 PM, Moore, Matthew J. <
> [EMAIL PROTECTED]> wrote:****
>
> Has anyone run Accumulo on a single server straight from memory?  Probably
> using something like a Fusion  IO drive.  We are trying to use it without
> using an SSD or any spinning discs.****
>
>  ****
>
> *Matthew Moore*****
>
> Systems Engineer****
>
> SAIC, ISBU****
>
> Columbia, MD****
>
> 410-312-2542****
>
>  ****
>
> ** **
>

Matthew,

I don't know of anyone who has done this, but I believe you could:
1. mount a RAM disk
2. point the hdfs core-site.xml fs.default.name property to file:///
3. point the accumulo-site.xml instance.dfs.dir property to a directory on
the RAM disk
4. disable the WAL for all tables by setting the accumulo-site.xml
table.walog.enabled to false
5. initialize and start up accumulo as you regularly would and cross your
fingers

Of course, the "you may lose data" and "this is not an officially supported
configuration" caveats apply. Out of curiosity, what would you be trying to
accomplish with this configuration?

Adam
On Tue, Sep 11, 2012 at 12:02 PM, Moore, Matthew J. <
[EMAIL PROTECTED]> wrote:

> Has anyone run Accumulo on a single server straight from memory?  Probably
> using something like a Fusion  IO drive.  We are trying to use it without
> using an SSD or any spinning discs.****
>
> ** **
>
> *Matthew Moore*****
>
> Systems Engineer****
>
> SAIC, ISBU****
>
> Columbia, MD****
>
> 410-312-2542****
>
> ** **
>

Adam,

It does look like we are the first to try this.  We are trying to keep
everything in memory and as a result there is no minor compactions, and
probably major compactions to make tables larger.  We tried this on SSDs
using a file system and we were not getting the processing speeds that
we had wanted.

 

Matt

 

 

From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]
On Behalf Of Adam Fuchs
Sent: Tuesday, September 11, 2012 5:30 PM
To: [EMAIL PROTECTED]
Subject: Re: Running Accumulo straight from Memory

 

Matthew,

 

I don't know of anyone who has done this, but I believe you could:

1. mount a RAM disk

2. point the hdfs core-site.xml fs.default.name
<http://fs.default.name/>  property to file:///

3. point the accumulo-site.xml instance.dfs.dir property to a directory
on the RAM disk

4. disable the WAL for all tables by setting the accumulo-site.xml
table.walog.enabled to false

5. initialize and start up accumulo as you regularly would and cross
your fingers

Of course, the "you may lose data" and "this is not an officially
supported configuration" caveats apply. Out of curiosity, what would you
be trying to accomplish with this configuration?

 

Adam

 

 

On Tue, Sep 11, 2012 at 12:02 PM, Moore, Matthew J.
<[EMAIL PROTECTED]> wrote:

Has anyone run Accumulo on a single server straight from memory?
Probably using something like a Fusion  IO drive.  We are trying to use
it without using an SSD or any spinning discs.

 

Matthew Moore

Systems Engineer

SAIC, ISBU

Columbia, MD

410-312-2542

 

 

Matt,

  Did you see Eric Newton's response yesterday? Running on a ram disk has been done; however minor and major compactions will still occur.

- Dave

----- Original Message -----
From: "Matthew J. Moore" <[EMAIL PROTECTED]>
To: [EMAIL PROTECTED]
Sent: Wednesday, September 12, 2012 12:32:31 PM
Subject: RE: Running Accumulo straight from Memory
Adam,

It does look like we are the first to try this.  We are trying to keep everything in memory and as a result there is no minor compactions, and probably major compactions to make tables larger.  We tried this on SSDs using a file system and we were not getting the processing speeds that we had wanted.

 

Matt

 

 
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of Adam Fuchs
Sent: Tuesday, September 11, 2012 5:30 PM
To: [EMAIL PROTECTED]
Subject: Re: Running Accumulo straight from Memory

 

Matthew,
 
I don't know of anyone who has done this, but I believe you could:
1. mount a RAM disk
2. point the hdfs core-site.xml  fs.default.name  property to file:///
3. point the accumulo-site.xml instance.dfs.dir property to a directory on the RAM disk
4. disable the WAL for all tables by setting the accumulo-site.xml table.walog.enabled to false
5. initialize and start up accumulo as you regularly would and cross your fingers

Of course, the "you may lose data" and "this is not an officially supported configuration" caveats apply. Out of curiosity, what would you be trying to accomplish with this configuration?
 
Adam
 

 
On Tue, Sep 11, 2012 at 12:02 PM, Moore, Matthew J. < [EMAIL PROTECTED] > wrote:

Has anyone run Accumulo on a single server straight from memory?  Probably using something like a Fusion  IO drive.  We are trying to use it without using an SSD or any spinning discs.

 

Matthew Moore

Systems Engineer

SAIC, ISBU

Columbia, MD

410-312-2542

 

 

Have you looked at in-memory specific software like Lucene, VoltDB,
MySQL? Doesn't Accumulo have too much overhead to act primarily as an
in-memory datastore? What advantage do you see Accumulo having?

Even if you are just using memory, minor and major compactions are
important to get compression, handle deletes, get sequential access (cache
line efficiency), use iterators, and introduce locality groups.

Adam
On Sep 12, 2012 12:33 PM, "Moore, Matthew J." <[EMAIL PROTECTED]>
wrote:

> Adam,****
>
> It does look like we are the first to try this.  We are trying to keep
> everything in memory and as a result there is no minor compactions, and
> probably major compactions to make tables larger.  We tried this on SSDs
> using a file system and we were not getting the processing speeds that we
> had wanted.****
>
> ** **
>
> Matt****
>
> ** **
>
> ** **
>
> *From:* [EMAIL PROTECTED][mailto:
> [EMAIL PROTECTED]] *On Behalf
> Of *Adam Fuchs
> *Sent:* Tuesday, September 11, 2012 5:30 PM
> *To:* [EMAIL PROTECTED]
> *Subject:* Re: Running Accumulo straight from Memory****
>
> ** **
>
> Matthew,****
>
> ** **
>
> I don't know of anyone who has done this, but I believe you could:****
>
> 1. mount a RAM disk****
>
> 2. point the hdfs core-site.xml fs.default.name property to file:///****
>
> 3. point the accumulo-site.xml instance.dfs.dir property to a directory on
> the RAM disk****
>
> 4. disable the WAL for all tables by setting the accumulo-site.xml
> table.walog.enabled to false****
>
> 5. initialize and start up accumulo as you regularly would and cross your
> fingers
>
> Of course, the "you may lose data" and "this is not an officially
> supported configuration" caveats apply. Out of curiosity, what would you be
> trying to accomplish with this configuration?****
>
> ** **
>
> Adam****
>
> ** **
>
> ** **
>
> On Tue, Sep 11, 2012 at 12:02 PM, Moore, Matthew J. <
> [EMAIL PROTECTED]> wrote:****
>
> Has anyone run Accumulo on a single server straight from memory?  Probably
> using something like a Fusion  IO drive.  We are trying to use it without
> using an SSD or any spinning discs.****
>
>  ****
>
> *Matthew Moore*****
>
> Systems Engineer****
>
> SAIC, ISBU****
>
> Columbia, MD****
>
> 410-312-2542****
>
>  ****
>
> ** **
>

Why would locality groups be useful in an in-memory system?

On Wed, Sep 12, 2012 at 4:53 PM, Adam Fuchs <[EMAIL PROTECTED]> wrote:
> Even if you are just using memory, minor and major compactions are important
> to get compression, handle deletes, get sequential access (cache line
> efficiency), use iterators, and introduce locality groups.

Yes, the effect of locality groups would be about the same in an in memory
system. The only exception would be if you're not using locality groups and
are fetching a particular column, the automatic seeking behavior of the
column filtering iterator would be more efficient with in memory rfiles.

Adam
On Sep 12, 2012 5:20 PM, "David Medinets" <[EMAIL PROTECTED]> wrote:

> Why would locality groups be useful in an in-memory system?
>
> On Wed, Sep 12, 2012 at 4:53 PM, Adam Fuchs <[EMAIL PROTECTED]> wrote:
> > Even if you are just using memory, minor and major compactions are
> important
> > to get compression, handle deletes, get sequential access (cache line
> > efficiency), use iterators, and introduce locality groups.
>

You guys are confirming what we predicted.  This was a "like to have"
from our customer and we wanted to see if anyone else had tried this.
Thanks.

 

Matt

 

 

 

From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]
On Behalf Of Adam Fuchs
Sent: Wednesday, September 12, 2012 5:28 PM
To: [EMAIL PROTECTED]
Subject: Re: Running Accumulo straight from Memory

 

Yes, the effect of locality groups would be about the same in an in
memory system. The only exception would be if you're not using locality
groups and are fetching a particular column, the automatic seeking
behavior of the column filtering iterator would be more efficient with
in memory rfiles.

Adam

On Sep 12, 2012 5:20 PM, "David Medinets" <[EMAIL PROTECTED]>
wrote:

Why would locality groups be useful in an in-memory system?

On Wed, Sep 12, 2012 at 4:53 PM, Adam Fuchs <[EMAIL PROTECTED]> wrote:
> Even if you are just using memory, minor and major compactions are
important
> to get compression, handle deletes, get sequential access (cache line
> efficiency), use iterators, and introduce locality groups.

On Wed, Sep 12, 2012 at 5:20 PM, David Medinets
<[EMAIL PROTECTED]> wrote:
> Why would locality groups be useful in an in-memory system?

Memory is fast, yet we still organize data in memory to make it really
fast (e.g. hash maps, sorted maps, bloom filters, etc)   Locality
groups are no different.  If using that data organization will make
what you are attempting to do faster, then you would probably use it.
Assume you have two locality groups and one contains 1% of your data
by volume and the other 99%.    Scanning just the locality group with
1% of the data will be faster than not having locality groups.  It
cuts down on the amount of data you have to read and processes from
memory.

>
> On Wed, Sep 12, 2012 at 4:53 PM, Adam Fuchs <[EMAIL PROTECTED]> wrote:
>> Even if you are just using memory, minor and major compactions are important
>> to get compression, handle deletes, get sequential access (cache line
>> efficiency), use iterators, and introduce locality groups.