[project-voldemort] Thing Cluster For Large Value

Hi
My cluster average put & get size is 90KB.  I am running on SSD based 
server and have 60 GB of jvm setting.  I setting db.cache.evictln�lse.
Is there any other setting that I need to change to optimize my puts ?
If anyone tried earlier and played with various settings?
Thanks
Mital
You received this message because you are subscribed to the Google Groups "project-voldemort" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
Visit this group at https://groups.google.com/group/project-voldemort.
For more options, visit https://groups.google.com/d/optout.

What is the latency you are getting for writes ? What version are you using
We generally run with bdb.cache.evictln=true to get better performance.
Please read through the following to see what other settings you can try on.
https://github.com/voldemort/voldemort/blob/master/bin/PREUPGRADE_FOR_1_1_X_README
Ability to move data off disk. This is very GC friendly, relying on OS page
   cache for the data and using the JVM heap only for index. This is
achieved
   by setting "bdb.cache.evictln" server parameter to "true"
impact
   on online traffic (Restore, Rebalance, Retention). This is achieved by
   setting "bdb.minimize.scan.impact" to "true"
cost
   of an additional delete()
Thanks,
Arun.
You received this message because you are subscribed to the Google Groups "project-voldemort" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
Visit this group at https://groups.google.com/group/project-voldemort.
For more options, visit https://groups.google.com/d/optout.

Mital,
Like Arun said, you should set bdb.cache.evictln=true. Setting it to false 
is an optimization for slow spinning disks at the cost of increased JVM 
heap usage (and consequently longer GC times), but you're on SSD. So, 
you'll actually have better performance if you set this to true.
Can you also give us more details on your read and write patterns?
- put() per second
- get() per second
- getAll() per second
- delete() per second
- For pu(), ratio of updates versus creates (overwrite versus new key)
- For getAll(), average and max number of keys per call
- Average and max key size
- Average and max value size
- How many stores on the cluster
You're best off making the JVM as small as possible, but you will need to 
have enough bdb cache to hold the hotset of the indexes of the stores.
A quick performance enhancement out of the box is to remove the 
ReadOnlyStorageEngineConfiguration from the storage.configs parameter. That 
way you only have the BdbStorageEngine running in your app.
This is one of our cluster configurations that hosts 50 stores and gets 
about a peak of 70,000 writes a second against billions of keys:
admin.max.threads@
bdb.cache.evictln=true
bdb.cache.size GB
bdb.checkpoint.interval.bytes!47483648
bdb.cleaner.interval.bytes728640
bdb.cleaner.lazy.migration�lse
bdb.cleaner.min.file.utilization=0
bdb.cleaner.threads=1
bdb.enable=true
bdb.evict.by.level=true
bdb.expose.space.utilization=true
bdb.lock.nLockTablesG
bdb.minimize.scan.impact=true
bdb.one.env.per.store=true
bdb.raw.property.string=je.cleaner.adjustUtilization�lse
data.directory=${voldemort.data.dir}
enable.server.routing�lse
enable.verbose.logging�lse
http.enable�lse
max.proxy.put.threadsP
nio.connector.selectorsP
num.scan.permits=2
restore.data.timeout.sec14000
retention.cleanup.first.start.hour=3
scheduler.threads$
slop.frequency.ms`0000
storage.configs=voldemort.store.bdb.BdbStorageConfiguration
stream.read.byte.per.sec 9715200
stream.write.byte.per.secx643200
voldemort.home=${voldemort.home.dir}
Some of the settings, 
like bdb.cleaner.threads, bdb.checkpoint.interval.bytes and 
bdb.cleaner.interval.bytes depend heavily on how frequently you create new 
keys and overwrite existing keys and how large the average and peak write 
sizes are.
We host that config in a 31gb Xms/Xmx JVM heap with UseCompressedOops set.
You received this message because you are subscribed to the Google Groups "project-voldemort" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
Visit this group at https://groups.google.com/group/project-voldemort.
For more options, visit https://groups.google.com/d/optout.

Greetings, I'm using the 2.4.x kernel, with the ext3 filesystem. I've talked to a bunch of people and everyone says 'apache doesnt have a file size limit' but they sure do! I have a 3 gig file (don't ask, why.. just go with the flow) and it gives me a permission deined error, when I try to view it. It also says in the error_log [Sun Mar 17 15:31:50 2002] [error] [client x.x.x.x] (75)Value too
Hi all, My system reports: Open_tables 512 Opened_tables 24,429 The docs say that if the latter is high I should increase the table cache size. (currently at 512) How does one decide what size to increase it to? And is there a problem with one of the applications that's making this figure so high? Or is this normal behaviour? OS: RH9 Dual 2.4 Xeon 1 GIG RAM (btw, this kind
Hey, Could someone give me some pointers how to use custom key-value serialization to store arbitrary data types? Maybe there is a documentation I am trying to find a Java client that supports this in a straight-forward way. Default Java serialization would also be fine. Thanks, Gyula You received this message because you are subscribed to the Google Groups "project-voldemort"
Apache/2.0.54 on Debian Stable. The file is 2.3GB (2,398,513,344) so I assume that the file is just too large for this build of Apache. In the log file I get: [Fri Sep 29 10:34:38 2006] [error] (75)Value too large for defined data type: access to /training/webcasts/webcast_data/161/webcast.mov failed But, there's a few odd things that happen I'm not clear on: 1) Google for that error message turns
Hi, how can I skip some value when build a ReadOnly store. In the mapper , I override makeKey and makeVlaue(), what to do if I don't want current value in the store output? public Object makeKey(NullWritable key, ProfileWritable value) { return null??? } return null seem create an exception. Thanks, You received this message because you are subscribed to the Google Groups "project-voldemort
Hi all, In one of our customers databases there is table that have an extremely high max_data_length of about 256TB(!). Another have 1 TB, while the rest is at 4GB. (all of these tables are basically the same, but with these two as the ones withthe most data). The largest table is currently at about 8.7 GB. Is there any practical consequences of having such overly large max_table_length
My configuration mysql 4.0.16 myodbc 3.51 My application is developed using visual c/c++ 6.0 and MFC I have a table that has an autoincrement primary key. I'm using the CRecorset classes to perform all data manipulation after the tables are created with the mysql client. At what appears to be a random occurance, when I insert a new record, the autoincrement value gets set to a very
0
I am writing a record, which has some array elements in it. In one of the records the number of entries for the array are around 600, in that case when i append a value in get an error "Value too large for file block size". And the worst part is that, once i hit this error for records after that i get the same error. I am using avro-c, the error comes while calling the routine "avro_file_writer_append_value
Hi, Previously we are using ActiveMQConnectionFactory alone, and we observed that while starting the JBoss Server ActiveMQ is creating Multiple Threads and closing( ActiveMQ Task 1, ActiveMQ Task 2,........), we thought it creating Multiple Connections/Session are creating and closing. As a resolution, we want to use CachingConnectionFactory, to configure CachingConnectionFactory we need