You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by China Stoffen <ch...@yahoo.com> on 2011/09/03 01:30:52 UTC
commodity server spec
Hi,
Is there any recommendation about commodity server hardware specs
if 100TB database size is expected and its heavily write application.
Should
I got with high powered CPU (12 cores) and 48TB HDD and 640GB RAM and
total of 3 servers of this spec. Or many smaller commodity servers are
recommended?
Thanks.
China
Re: commodity server spec
Posted by Bill <bi...@dehora.net>.
Mongodb, last time I looked does not scale horizontally.
I've seen reasonable behavour putting Cassandra database tables onto
remote filers, but you absolutely have to test against the SAN
configuration and carefully manage things like concurrent reader/writer
settings, the fs and cassandra caches, etc. You generally won't be
recommended to use a NAS/SAN for this class of system.
The commitlogs work best on attached (dedicated) disk.
Bill
On 04/09/11 14:08, China Stoffen wrote:
> Then what will be the sweetspot for Cassandra? I am more interested in
> Cassandra because my application is write heavy.
>
> Till now what I have understood is that Cassandra will not work best for
> SANs too?
>
> P.S
> Mongodb is also a nosql database and designed for horizontal scaling
> then how its good for the same hardware for which Cassandra is not a
> good candidate?
>
>
> ----- Original Message -----
> From: Bill <bi...@dehora.net>
> To: user@cassandra.apache.org
> Cc:
> Sent: Sunday, September 4, 2011 4:34 AM
> Subject: Re: commodity server spec
>
> [100% agree with Chris]
>
> China, the machines you're describing sound nice for
> mongodb/postgres/mysql, but probably not the sweetspot for Cassandra.
>
> Obviously (well depending on near term load) you don't want to get
> burned on excess footprint. But a realistic, don't lose data, be fairly
> available deployment is going to span at least 2 racks/power supplies
> and have data replicated offsite (at least as passive for DR). So I
> would consider 6-9 relatively weaker servers rather than 3 scale up
> joints. You'll save some capex, and the amount of opex overhead is
> probably worth it traded off against the operational risk. 3 is an
> awkward number to operate for anything that needs to be available
> (although many people seem to start with that, I am guessing because
> triplication is traditionally understood under failure) as it
> immediately puts 50% extra load on the remaining 2 when one node goes
> away. One will go away, even transiently, when it is upgraded, crashes,
> gets into a funk due to compaction or garbage collection, and load will
> then be shunted onto the other 2 - remember Cassandra has no
> backoff/throttling in place. I'd allow for something breaking at some
> point (dbs even the mature ones, fail from time to time) and 2 doesn't
> give you much room to maneuver in production.
>
> Bill
>
>
> On 03/09/11 23:05, Chris Goffinet wrote:
> > It will also depend on how long you can handle recovery time. So imagine
> > this case:
> >
> > 3 nodes w/ RF of 3
> > Each node has 30TB of space used (you never want to fill up entire node).
> > If one node fails and you must recover, that will take over 3.6 days in
> > just transferring data alone. That's with a sustained 800megabit/s
> > (100MB/s). In the real world it's going to fluctuate so add some
> > padding. Also, since you will be saturating one of the other nodes, now
> > you're network latency performance is suffering and you only have 1
> > machine to handle the remaining traffic while you're recovering. And if
> > you want to expand the cluster in the future (more nodes), the amount of
> > data to transfer is going to be very large and most likely days to add
> > machines. From my experience it's must better to have a larger cluster
> > setup upfront for future growth than getting by with 6-12 nodes at the
> > start. You will feel less pain, easier to manage node failures (bad
> > disks, mem, etc).
> >
> > 3 nodes with RF of 1 wouldn't make sense.
> >
> >
> > On Sat, Sep 3, 2011 at 4:05 AM, China Stoffen <chinastoffen@yahoo.com
> <ma...@yahoo.com>
> > <mailto:chinastoffen@yahoo.com <ma...@yahoo.com>>> wrote:
> >
> > Many small servers would drive up the hosting cost way too high so
> > want to avoid this solution if we can.
> >
> > ----- Original Message -----
> > From: Radim Kolar <hsn@sendmail.cz <ma...@sendmail.cz>
> <mailto:hsn@sendmail.cz <ma...@sendmail.cz>>>
> > To: user@cassandra.apache.org <ma...@cassandra.apache.org>
> <mailto:user@cassandra.apache.org <ma...@cassandra.apache.org>>
> > Cc:
> > Sent: Saturday, September 3, 2011 9:37 AM
> > Subject: Re: commodity server spec
> >
> > many smaller servers way better
> >
> >
>
Re: commodity server spec
Posted by China Stoffen <ch...@yahoo.com>.
Then what will be the sweetspot for Cassandra? I am more interested in Cassandra because my application is write heavy.
Till now what I have understood is that Cassandra will not work best for SANs too?
P.S
Mongodb is also a nosql database and designed for horizontal scaling then how its good for the same hardware for which Cassandra is not a good candidate?
----- Original Message -----
From: Bill <bi...@dehora.net>
To: user@cassandra.apache.org
Cc:
Sent: Sunday, September 4, 2011 4:34 AM
Subject: Re: commodity server spec
[100% agree with Chris]
China, the machines you're describing sound nice for mongodb/postgres/mysql, but probably not the sweetspot for Cassandra.
Obviously (well depending on near term load) you don't want to get burned on excess footprint. But a realistic, don't lose data, be fairly available deployment is going to span at least 2 racks/power supplies and have data replicated offsite (at least as passive for DR). So I would consider 6-9 relatively weaker servers rather than 3 scale up joints. You'll save some capex, and the amount of opex overhead is probably worth it traded off against the operational risk. 3 is an awkward number to operate for anything that needs to be available (although many people seem to start with that, I am guessing because triplication is traditionally understood under failure) as it immediately puts 50% extra load on the remaining 2 when one node goes away. One will go away, even transiently, when it is upgraded, crashes, gets into a funk due to compaction or garbage collection, and load will then be shunted onto the other 2 - remember Cassandra has no
backoff/throttling in place. I'd allow for something breaking at some point (dbs even the mature ones, fail from time to time) and 2 doesn't give you much room to maneuver in production.
Bill
On 03/09/11 23:05, Chris Goffinet wrote:
> It will also depend on how long you can handle recovery time. So imagine
> this case:
>
> 3 nodes w/ RF of 3
> Each node has 30TB of space used (you never want to fill up entire node).
> If one node fails and you must recover, that will take over 3.6 days in
> just transferring data alone. That's with a sustained 800megabit/s
> (100MB/s). In the real world it's going to fluctuate so add some
> padding. Also, since you will be saturating one of the other nodes, now
> you're network latency performance is suffering and you only have 1
> machine to handle the remaining traffic while you're recovering. And if
> you want to expand the cluster in the future (more nodes), the amount of
> data to transfer is going to be very large and most likely days to add
> machines. From my experience it's must better to have a larger cluster
> setup upfront for future growth than getting by with 6-12 nodes at the
> start. You will feel less pain, easier to manage node failures (bad
> disks, mem, etc).
>
> 3 nodes with RF of 1 wouldn't make sense.
>
>
> On Sat, Sep 3, 2011 at 4:05 AM, China Stoffen <chinastoffen@yahoo.com
> <ma...@yahoo.com>> wrote:
>
> Many small servers would drive up the hosting cost way too high so
> want to avoid this solution if we can.
>
> ----- Original Message -----
> From: Radim Kolar <hsn@sendmail.cz <ma...@sendmail.cz>>
> To: user@cassandra.apache.org <ma...@cassandra.apache.org>
> Cc:
> Sent: Saturday, September 3, 2011 9:37 AM
> Subject: Re: commodity server spec
>
> many smaller servers way better
>
>
Re: commodity server spec
Posted by Bill <bi...@dehora.net>.
[100% agree with Chris]
China, the machines you're describing sound nice for
mongodb/postgres/mysql, but probably not the sweetspot for Cassandra.
Obviously (well depending on near term load) you don't want to get
burned on excess footprint. But a realistic, don't lose data, be fairly
available deployment is going to span at least 2 racks/power supplies
and have data replicated offsite (at least as passive for DR). So I
would consider 6-9 relatively weaker servers rather than 3 scale up
joints. You'll save some capex, and the amount of opex overhead is
probably worth it traded off against the operational risk. 3 is an
awkward number to operate for anything that needs to be available
(although many people seem to start with that, I am guessing because
triplication is traditionally understood under failure) as it
immediately puts 50% extra load on the remaining 2 when one node goes
away. One will go away, even transiently, when it is upgraded, crashes,
gets into a funk due to compaction or garbage collection, and load will
then be shunted onto the other 2 - remember Cassandra has no
backoff/throttling in place. I'd allow for something breaking at some
point (dbs even the mature ones, fail from time to time) and 2 doesn't
give you much room to maneuver in production.
Bill
On 03/09/11 23:05, Chris Goffinet wrote:
> It will also depend on how long you can handle recovery time. So imagine
> this case:
>
> 3 nodes w/ RF of 3
> Each node has 30TB of space used (you never want to fill up entire node).
> If one node fails and you must recover, that will take over 3.6 days in
> just transferring data alone. That's with a sustained 800megabit/s
> (100MB/s). In the real world it's going to fluctuate so add some
> padding. Also, since you will be saturating one of the other nodes, now
> you're network latency performance is suffering and you only have 1
> machine to handle the remaining traffic while you're recovering. And if
> you want to expand the cluster in the future (more nodes), the amount of
> data to transfer is going to be very large and most likely days to add
> machines. From my experience it's must better to have a larger cluster
> setup upfront for future growth than getting by with 6-12 nodes at the
> start. You will feel less pain, easier to manage node failures (bad
> disks, mem, etc).
>
> 3 nodes with RF of 1 wouldn't make sense.
>
>
> On Sat, Sep 3, 2011 at 4:05 AM, China Stoffen <chinastoffen@yahoo.com
> <ma...@yahoo.com>> wrote:
>
> Many small servers would drive up the hosting cost way too high so
> want to avoid this solution if we can.
>
> ----- Original Message -----
> From: Radim Kolar <hsn@sendmail.cz <ma...@sendmail.cz>>
> To: user@cassandra.apache.org <ma...@cassandra.apache.org>
> Cc:
> Sent: Saturday, September 3, 2011 9:37 AM
> Subject: Re: commodity server spec
>
> many smaller servers way better
>
>
Re: commodity server spec
Posted by Chris Goffinet <cg...@chrisgoffinet.com>.
It will also depend on how long you can handle recovery time. So imagine
this case:
3 nodes w/ RF of 3
Each node has 30TB of space used (you never want to fill up entire node).
If one node fails and you must recover, that will take over 3.6 days in
just transferring data alone. That's with a sustained 800megabit/s
(100MB/s). In the real world it's going to fluctuate so add some padding.
Also, since you will be saturating one of the other nodes, now you're
network latency performance is suffering and you only have 1 machine to
handle the remaining traffic while you're recovering. And if you want to
expand the cluster in the future (more nodes), the amount of data to
transfer is going to be very large and most likely days to add machines.
>From my experience it's must better to have a larger cluster setup upfront
for future growth than getting by with 6-12 nodes at the start. You will
feel less pain, easier to manage node failures (bad disks, mem, etc).
3 nodes with RF of 1 wouldn't make sense.
On Sat, Sep 3, 2011 at 4:05 AM, China Stoffen <ch...@yahoo.com>wrote:
> Many small servers would drive up the hosting cost way too high so want to
> avoid this solution if we can.
>
> ----- Original Message -----
> From: Radim Kolar <hs...@sendmail.cz>
> To: user@cassandra.apache.org
> Cc:
> Sent: Saturday, September 3, 2011 9:37 AM
> Subject: Re: commodity server spec
>
> many smaller servers way better
>
Re: commodity server spec
Posted by China Stoffen <ch...@yahoo.com>.
Many small servers would drive up the hosting cost way too high so want to avoid this solution if we can.
----- Original Message -----
From: Radim Kolar <hs...@sendmail.cz>
To: user@cassandra.apache.org
Cc:
Sent: Saturday, September 3, 2011 9:37 AM
Subject: Re: commodity server spec
many smaller servers way better
Re: commodity server spec
Posted by Radim Kolar <hs...@sendmail.cz>.
many smaller servers way better
Re: commodity server spec
Posted by China Stoffen <ch...@yahoo.com>.
>>In general, more smaller is better than fewer big. Probably go for
>>what's cost-effective.
Cost effective solution is few and fat servers because it also saves hosting cost.
>>The exception to that would be if you're truly only caring about
>>writes and have *very* few reads that are not latency critical (so
>>you're okay with waiting for several disk seeks on reads and the
>>number of reads is low enough that serving them from platters will
>>work). In such cases it might make sense to have fewer Big Fat
>>Machines with lots of memory and a lot of disk space. But... even so.
>>I would not recommend huge 48 tb nodes... unless you really know what
>>you're doing.
I want writes should be as fast as possible but reads are not necessary to be in milliseconds.
If you don't recommend 48tb then how much max disk space I can go with?
----- Original Message -----
From: Peter Schuller <pe...@infidyne.com>
To: user@cassandra.apache.org; China Stoffen <ch...@yahoo.com>
Cc:
Sent: Saturday, September 3, 2011 1:08 PM
Subject: Re: commodity server spec
> Is there any recommendation about commodity server hardware specs if 100TB
> database size is expected and its heavily write application.
> Should I got with high powered CPU (12 cores) and 48TB HDD and 640GB RAM and
> total of 3 servers of this spec. Or many smaller commodity servers are
> recommended?
In general, more smaller is better than fewer big. Probably go for
what's cost-effective.
In your case, 100 TB is *quite* big. I would definitely recommend
against doing anything like your 3 server setup. You'll probably want
100-1000 small servers.
The exception to that would be if you're truly only caring about
writes and have *very* few reads that are not latency critical (so
you're okay with waiting for several disk seeks on reads and the
number of reads is low enough that serving them from platters will
work). In such cases it might make sense to have fewer Big Fat
Machines with lots of memory and a lot of disk space. But... even so.
I would not recommend huge 48 tb nodes... unless you really know what
you're doing.
In reality, more information about your use-case would be required to
offer terribly useful advice.
--
/ Peter Schuller (@scode on twitter)
Re: commodity server spec
Posted by Peter Schuller <pe...@infidyne.com>.
> Is there any recommendation about commodity server hardware specs if 100TB
> database size is expected and its heavily write application.
> Should I got with high powered CPU (12 cores) and 48TB HDD and 640GB RAM and
> total of 3 servers of this spec. Or many smaller commodity servers are
> recommended?
In general, more smaller is better than fewer big. Probably go for
what's cost-effective.
In your case, 100 TB is *quite* big. I would definitely recommend
against doing anything like your 3 server setup. You'll probably want
100-1000 small servers.
The exception to that would be if you're truly only caring about
writes and have *very* few reads that are not latency critical (so
you're okay with waiting for several disk seeks on reads and the
number of reads is low enough that serving them from platters will
work). In such cases it might make sense to have fewer Big Fat
Machines with lots of memory and a lot of disk space. But... even so.
I would not recommend huge 48 tb nodes... unless you really know what
you're doing.
In reality, more information about your use-case would be required to
offer terribly useful advice.
--
/ Peter Schuller (@scode on twitter)