and, why did you set "JVM has 8G heap"?
8g, seems too big.

在 2010年6月2日 上午11:20,lwl <lwl.roger@gmail.com>写道:
3.32 concurrent read & 128 write in storage-conf.xml, other cache
enlarged as well.
--------------------

maybe you can try to decrease the size of caches.

在 2010年6月2日 上午11:14,Shuai Yuan <yuanshuai@supertool.net.cn>写道:

在 2010-06-02三的 10:37 +0800,lwl写道:
> is all the 4 servers' MEM  almost 100%?

Yes

> 在 2010年6月2日 上午10:12,Shuai Yuan <yuanshuai@supertool.net.cn>写
> 道:
>         Thanks lwl.
>
>         Then is there anyway of tuning this, faster flush to disk or
>         else?
>
>         Cheers,
>
>         Kevin
>
>         在 2010-06-02三的 09:57 +0800,lwl写道:
>
>         > MEM: almost 100% (16GB)
>         > -----------------
>         > maybe this is the bottleneck.
>         > writing concerns Memtable and SSTable in memory.
>         >
>         > 在 2010年6月2日 上午9:48,Shuai Yuan
>         <yuanshuai@supertool.net.cn>写
>         > 道:
>         >         在 2010-06-01二的 15:00 -0500,Jonathan Shook写道:
>         >         > Also, what are you meaning specifically by 'slow'?
>         Which
>         >         measurements
>         >         > are you looking at. What are your baseline
>         constraints for
>         >         your test
>         >         > system?
>         >         >
>         >
>         >         Actually, the problem is the utilizaton of
>         resources(for a
>         >         single
>         >         machine):
>         >         CPU: 700% / 1600% (16 cores)
>         >         MEM: almost 100% (16GB)
>         >         Swap: almost 0%
>         >         Disk IO(write): 20~30MB / 200MB (7.2k raid5,
>         benchmarked
>         >         previously)
>         >         NET: up to 100Mbps / 950Mbps (1Gbps, tuned and
>         benchmarked
>         >         previously)
>         >
>         >         So the speed of generating load, about 15M/s as
>         reported
>         >         before seems
>         >         quite slow to me. I assume the system should get at
>         least
>         >         about 50MB/s
>         >         of Disk IO speed.
>         >
>         >         MEM? I don't think it plays a major role in this
>         writing game.
>         >         What's
>         >         the bottleneck of the system?
>         >
>         >         P.S
>         >         about Consistency Level, I've tried ONE/DCQUORUM and
>         found ONE
>         >         is about
>         >         10-15% faster. However that's neither a promising
>         result.
>         >
>         >         Thanks!
>         >
>         >         Kevin
>         >
>         >         >
>         >         > 2010/6/1 史英杰 <shiyingjie1983@gmail.com>:
>         >         > > Hi, It would be better if we know which
>         Consistency Level
>         >         did you choose,
>         >         > > and what is the schema of test data?
>         >         > >
>         >         > > 在 2010年6月1日 下午4:48,Shuai Yuan
>         >         <yuanshuai@supertool.net.cn>写道:
>         >         > >>
>         >         > >> Hi all,
>         >         > >>
>         >         > >> I'm testing writing speed of cassandra with 4
>         servers.
>         >         I'm confused by
>         >         > >> the behavior of cassandra.
>         >         > >>
>         >         > >> ---env---
>         >         > >> load-data app written in c++, using
>         libcassandra (w/
>         >         modified batch
>         >         > >> insert)
>         >         > >> 20 writing threads in 2 processes running on 2
>         servers
>         >         > >>
>         >         > >> ---optimization---
>         >         > >> 1.turn log level to INFO
>         >         > >> 2.JVM has 8G heap
>         >         > >> 3.32 concurrent read & 128 write in
>         storage-conf.xml,
>         >         other cache
>         >         > >> enlarged as well.
>         >         > >>
>         >         > >> ---result---
>         >         > >> 1-monitoring by `date;nodetool -h host ring`
>         >         > >> I add all load together and measure the writing
>         speed by
>         >         > >> (load_difference / time_difference), and I get
>         about
>         >         15MB/s for the
>         >         > >> whole cluster.
>         >         > >>
>         >         > >> 2-monitoring by `iostat -m 10`
>         >         > >> I can watch the disk_io from the system level
>         and have
>         >         about 10MB/s -
>         >         > >> 65MB/s for a single machine. Very big variance
>         over time.
>         >         > >>
>         >         > >> 3-monitoring by `iptraf -g`
>         >         > >> In this way I watch the communication between
>         servers and
>         >         get about
>         >         > >> 10MB/s for a single machine.
>         >         > >>
>         >         > >> ---opinion---
>         >         > >> So, have you checked the writing speed of
>         cassandra? I
>         >         feel it's quite
>         >         > >> slow currently.
>         >         > >>
>         >         > >> Could anyone confirm this is the normal writing
>         speed of
>         >         cassandra, or
>         >         > >> please provide someway of improving it?
>         >         > >> --
>         >         > >> Kevin Yuan
>         >         > >> www.yuan-shuai.info
>         >         > >>
>         >         > >>
>         >         > >
>         >         > >
>         >         >
>         >
>         >
>         >         --
>         >
>         >         Kevin Yuan
>         >         www.yuan-shuai.info
>         >
>         >
>         >
>         >
>         >
>
>
>         --
>         Shuai Yuan 袁帅
>         Supertool Corp. 北京学之途网络科技有限公司
>
>         www.yuan-shuai.info
>
>
>

--
Shuai Yuan 袁帅
Supertool Corp. 北京学之途网络科技有限公司
www.yuan-shuai.info