Home | About | Sematext search-lucene.com search-hadoop.com
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB
 Search Hadoop and all its subprojects:

Switch to Threaded View
HBase >> mail # user >> High IPC Latency


Copy link to this message
-
Re: High IPC Latency
Can you reproduce this against a single, local region server?
Any chance that you can try with the just released 0.94.2?
I would love to debug this. If would be a tremendous help if you had a little test program that reproduces this against a single server, so that I can see what is going on.

Thanks.

-- Lars

----- Original Message -----
From: Yousuf Ahmad <[EMAIL PROTECTED]>
To: [EMAIL PROTECTED]; lars hofhansl <[EMAIL PROTECTED]>
Cc: Ivan Brondino <[EMAIL PROTECTED]>; Ricardo Vilaça <[EMAIL PROTECTED]>
Sent: Thursday, October 18, 2012 12:59 PM
Subject: Re: High IPC Latency

Hi,

Thank you for your questions guys.

We are using HBase 0.92 with HDFS 1.0.1.

The experiment lasts 15 minutes. The measurements stabilize in the first
two minutes of the run.

The data is distributed almost evenly across the regionservers so each
client hits most of them over the course of the experiment. However, for
the data we have, any given multi-get or scan should touch only one or at
most two regions.

The client caches the locations of the regionservers, so after a couple of
minutes of the experiment running, it wouldn't need to re-visit ZooKeeper,
I believe. Correct me if I am wrong please.

Regards,
Yousuf
On Thu, Oct 18, 2012 at 2:42 PM, lars hofhansl <[EMAIL PROTECTED]> wrote:

> Also, what version of HBase/HDFS is this using?
>
>
>
>
> ----- Original Message -----
> From: "Pamecha, Abhishek" <[EMAIL PROTECTED]>
> To: "[EMAIL PROTECTED]" <[EMAIL PROTECTED]>
> Cc: Ivan Brondino <[EMAIL PROTECTED]>; Ricardo Vilaça <
> [EMAIL PROTECTED]>
> Sent: Thursday, October 18, 2012 11:38 AM
> Subject: RE: High IPC Latency
>
> Is it sustained for the same client hitting the same region server OR does
> it get better for the same client-RS combination when run for longer
> duration?  Trying to eliminate Zookeeper from this.
>
> Thanks,
> Abhishek
>
> From: Yousuf Ahmad [mailto:[EMAIL PROTECTED]]
> Sent: Thursday, October 18, 2012 11:26 AM
> To: [EMAIL PROTECTED]
> Cc: Ivan Brondino; Ricardo Vilaça
> Subject: High IPC Latency
>
> Hello,
>
> We are seeing slow times for read operations in our experiments. We are
> hoping that you guys can help us figure out what's going wrong.
>
> Here are some details:
>
>   *   We are running a read-only benchmark on our HBase cluster.
>   *
>   *   There are 10 regionservers, each co-located with a datanode. HDFS
> replication is 3x.
>   *   All the data read by the experiment is already in the block cache
> and the hit ratio is 99%.
>   *
>   *   We have 10 clients, each with around 400 threads making a mix of
> read-only requests involving multi-gets and scans.
>   *
>   *   We settled on the default client pool type/size (roundrobin/1) and a
> regionserver handler count of 100 after testing various combinations to see
> what setting worked best.
>   *
>   *   Our scans are short, fetching around 10 rows on average. Scanner
> caching is set to 50.
>   *   An average row in a scan has either around 10 columns (small row) or
> around 200 columns (big row).
>   *
>   *   Our multi-gets fetch around 200 rows on average.
>   *   An average row in a multi-get has around 10 columns.
>   *   Each column holds an integer (encoded into bytes).
>   *
>   *   None of the machines involved reach CPU, memory, or IO saturation.
> In fact resource utilization stays quite low.
>   *
>   *   Our statistics show that the average time for a scan, measured
> starting from the first scanner.next() call to the last one which returns a
> null, is around 2-3 seconds.
>   *   Since we use scanner caching, the major portion of this time (around
> 2 seconds) is spent on the first call to next(), while the remaining calls
> take a negligible amount of time.
>   *   Similarly, we see that a multi-get on average takes around 2 seconds.
>   *   A single get on average takes around 1 second.
> We are not sure what the bottleneck is or where it lies. We thought we
> should look deeper into what is going on at the regionservers. We monitored
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB