You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jcs-users@jakarta.apache.org by Randy Watler <wa...@wispertel.net> on 2006/11/03 16:42:43 UTC

JDBCDiskCache as Remote Cache

We are attempting to use a JDBCDiskCache in a shared mode with multiple
cache clients using it as their backing store. We are also interested in
using the TCPLateralCache in "remove on put" mode to help keep the
clients roughly in sync where their extents overlap, (albeit not at a
transactional level). We are not using a remote cache because each cache
client will have a largely orthogonal working set in relation to the
other clients and that would force the single remote cache to hold a
very large working set for no advantage. Conceptually speaking, we are
attempting to use the JDBCDiskCache as a remote cache instead.

Therein lies a problem though. Because the JDBCDiskCache is in fact a
disk cache, JCS propagates lateral updates as removes to this auxiliary
cache. This ends up actually removing the elements from the
JDBCDiskCache instead of simply removing the items from the memory cache
as would happen if it were a remote cache.

I am wondering if it would be reasonable to allow a JDBCDiskCache to
masquerade as a remote cache by indicating that its cache type is
remote? This would seemingly allow the cache to operate as a "remote"
disk cache. True?

Any comments/tips would be appreciated. Thanks,

Randy



---------------------------------------------------------------------
To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jcs-users-help@jakarta.apache.org


Re: JDBCDiskCache as Remote Cache

Posted by Randy Watler <wa...@wispertel.net>.
Aaron,

Thank you very much for responding.

I am certainly not avoiding the remote server implementation, but I am
concerned about how it will scale in our case. We have a very large
working set and the individual lateral cache clients hold data unique to
them. For example, say I have 5 cache clients that share only 30% of
their cached working set. I believe then that if X is the given cache
size, the remote server would have to be sized 30% + 5 * 60%, or 330% of
X. Given this, I would rather just have the 5 cache clients w/o a remote
cache. Let me know if this logic is flawed.

I was thinking along the same lines as you with a "shared" type. Clearly
the original disk cache is not sharable, but the JDBC disk cache should
be.... and that is what I am trying to do :-).

Randy

On Fri, 2006-11-03 at 08:57 -0800, Aaron Smuts wrote:
> The remote cache doesn't work the way you described
> exactly.  If a remove request goes to the remote
> server, it sends it to all listeners and to all
> auxiliaries.  There is no command to remove from
> memory only.  
> 
> Just like the lateral cache, it can be configured to
> send remove requests.  If you have the lateral client
> configured to get from the other laterals, then the
> laterals start to look like a bunch of remote caches. 
> However, this is not advisable when  you have a buch
> of machines.  You don't want to have to search 50
> boxes for every cache miss. . . . .  
> 
> If you are using a shared jdbc disk cache, then there
> is no reason not to use the remote server.  I push
> over 500 puts a second through a remote cluster and do
> the same number of gets . . .  If you are worried
> about memory, then configure the remote server to not
> store very much or anything in memory.  That's how I
> use it for most regions.  Use the remote cache as a
> way to access the JDBC disk cache.  Also, the remote
> will take care of the invalidation messages for you. 
> It is the most scallable way to go.  Also, you can
> partition the data into multiple rmeote clusters in
> the future if you needed to scale.   . . .
> 
> Hmmn.  I see what you are trying to do.  If the disk
> cache was configured as a non-local auxiliary (i.e.
> remote or lateral) then yes, the remove would not
> remove the item from disk.  What we need is a new type
> called shared or remote disk.  This would do the
> trick.        The assumption that all disk caches are
> local is false in this case.  Perhaps we could use the
> disk mode setting.   . . . Let me think about it a bit
> more.  For now, I suggest that you use the remote
> cache.
> 
> Cheers,
> 
> Aaron
>  
> 
> --- Randy Watler <wa...@wispertel.net> wrote:
> 
> > We are attempting to use a JDBCDiskCache in a shared
> > mode with multiple
> > cache clients using it as their backing store. We
> > are also interested in
> > using the TCPLateralCache in "remove on put" mode to
> > help keep the
> > clients roughly in sync where their extents overlap,
> > (albeit not at a
> > transactional level). We are not using a remote
> > cache because each cache
> > client will have a largely orthogonal working set in
> > relation to the
> > other clients and that would force the single remote
> > cache to hold a
> > very large working set for no advantage.
> > Conceptually speaking, we are
> > attempting to use the JDBCDiskCache as a remote
> > cache instead.
> > 
> > Therein lies a problem though. Because the
> > JDBCDiskCache is in fact a
> > disk cache, JCS propagates lateral updates as
> > removes to this auxiliary
> > cache. This ends up actually removing the elements
> > from the
> > JDBCDiskCache instead of simply removing the items
> > from the memory cache
> > as would happen if it were a remote cache.
> > 
> > I am wondering if it would be reasonable to allow a
> > JDBCDiskCache to
> > masquerade as a remote cache by indicating that its
> > cache type is
> > remote? This would seemingly allow the cache to
> > operate as a "remote"
> > disk cache. True?
> > 
> > Any comments/tips would be appreciated. Thanks,
> > 
> > Randy
> > 
> > 
> > 
> >
> ---------------------------------------------------------------------
> > To unsubscribe, e-mail:
> > jcs-users-unsubscribe@jakarta.apache.org
> > For additional commands, e-mail:
> > jcs-users-help@jakarta.apache.org
> > 
> > 
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jcs-users-help@jakarta.apache.org
> 
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jcs-users-help@jakarta.apache.org


Re: JDBCDiskCache as Remote Cache

Posted by Randy Watler <wa...@wispertel.net>.
Aaron,

Just a quick note... overriding JDBCDiskCache.getCacheType() seems to be
working well for us so far. Thanks again for the input. I will let you
know if I run into any unexpected features.

Randy

On Fri, 2006-11-03 at 11:20 -0700, Randy Watler wrote:
> Aaron,
> 
> Cool. I was going to try this out today after reviewing the
> ramifications in the code of claiming the JDBCDiskCache was remote. It
> is encouraging to hear that you think this might work as well!
> 
> Thanks for the continued feedback.
> 
> Randy
> 
> On Fri, 2006-11-03 at 09:11 -0800, Aaron Smuts wrote:
> > You could of course, create a wrapper for the jdbc
> > disk cache that returned the remote type.  Just extend
> > it and override that method.  That would be a hack,
> > but it's easy since JCS is pluggable.
> > 
> > --- Aaron Smuts <as...@yahoo.com> wrote:
> > 
> > > The remote cache doesn't work the way you described
> > > exactly.  If a remove request goes to the remote
> > > server, it sends it to all listeners and to all
> > > auxiliaries.  There is no command to remove from
> > > memory only.  
> > > 
> > > Just like the lateral cache, it can be configured to
> > > send remove requests.  If you have the lateral
> > > client
> > > configured to get from the other laterals, then the
> > > laterals start to look like a bunch of remote
> > > caches. 
> > > However, this is not advisable when  you have a buch
> > > of machines.  You don't want to have to search 50
> > > boxes for every cache miss. . . . .  
> > > 
> > > If you are using a shared jdbc disk cache, then
> > > there
> > > is no reason not to use the remote server.  I push
> > > over 500 puts a second through a remote cluster and
> > > do
> > > the same number of gets . . .  If you are worried
> > > about memory, then configure the remote server to
> > > not
> > > store very much or anything in memory.  That's how I
> > > use it for most regions.  Use the remote cache as a
> > > way to access the JDBC disk cache.  Also, the remote
> > > will take care of the invalidation messages for you.
> > > 
> > > It is the most scallable way to go.  Also, you can
> > > partition the data into multiple rmeote clusters in
> > > the future if you needed to scale.   . . .
> > > 
> > > Hmmn.  I see what you are trying to do.  If the disk
> > > cache was configured as a non-local auxiliary (i.e.
> > > remote or lateral) then yes, the remove would not
> > > remove the item from disk.  What we need is a new
> > > type
> > > called shared or remote disk.  This would do the
> > > trick.        The assumption that all disk caches
> > > are
> > > local is false in this case.  Perhaps we could use
> > > the
> > > disk mode setting.   . . . Let me think about it a
> > > bit
> > > more.  For now, I suggest that you use the remote
> > > cache.
> > > 
> > > Cheers,
> > > 
> > > Aaron
> > >  
> > > 
> > > --- Randy Watler <wa...@wispertel.net> wrote:
> > > 
> > > > We are attempting to use a JDBCDiskCache in a
> > > shared
> > > > mode with multiple
> > > > cache clients using it as their backing store. We
> > > > are also interested in
> > > > using the TCPLateralCache in "remove on put" mode
> > > to
> > > > help keep the
> > > > clients roughly in sync where their extents
> > > overlap,
> > > > (albeit not at a
> > > > transactional level). We are not using a remote
> > > > cache because each cache
> > > > client will have a largely orthogonal working set
> > > in
> > > > relation to the
> > > > other clients and that would force the single
> > > remote
> > > > cache to hold a
> > > > very large working set for no advantage.
> > > > Conceptually speaking, we are
> > > > attempting to use the JDBCDiskCache as a remote
> > > > cache instead.
> > > > 
> > > > Therein lies a problem though. Because the
> > > > JDBCDiskCache is in fact a
> > > > disk cache, JCS propagates lateral updates as
> > > > removes to this auxiliary
> > > > cache. This ends up actually removing the elements
> > > > from the
> > > > JDBCDiskCache instead of simply removing the items
> > > > from the memory cache
> > > > as would happen if it were a remote cache.
> > > > 
> > > > I am wondering if it would be reasonable to allow
> > > a
> > > > JDBCDiskCache to
> > > > masquerade as a remote cache by indicating that
> > > its
> > > > cache type is
> > > > remote? This would seemingly allow the cache to
> > > > operate as a "remote"
> > > > disk cache. True?
> > > > 
> > > > Any comments/tips would be appreciated. Thanks,
> > > > 
> > > > Randy
> > > > 
> > > > 
> > > > 
> > > >
> > >
> > ---------------------------------------------------------------------
> > > > To unsubscribe, e-mail:
> > > > jcs-users-unsubscribe@jakarta.apache.org
> > > > For additional commands, e-mail:
> > > > jcs-users-help@jakarta.apache.org
> > > > 
> > > > 
> > > 
> > > 
> > >
> > ---------------------------------------------------------------------
> > > To unsubscribe, e-mail:
> > > jcs-users-unsubscribe@jakarta.apache.org
> > > For additional commands, e-mail:
> > > jcs-users-help@jakarta.apache.org
> > > 
> > > 
> > 
> > 
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
> > For additional commands, e-mail: jcs-users-help@jakarta.apache.org
> > 
> > 
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jcs-users-help@jakarta.apache.org
> 
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jcs-users-help@jakarta.apache.org


Re: JDBCDiskCache as Remote Cache

Posted by Randy Watler <wa...@wispertel.net>.
Aaron,

Cool. I was going to try this out today after reviewing the
ramifications in the code of claiming the JDBCDiskCache was remote. It
is encouraging to hear that you think this might work as well!

Thanks for the continued feedback.

Randy

On Fri, 2006-11-03 at 09:11 -0800, Aaron Smuts wrote:
> You could of course, create a wrapper for the jdbc
> disk cache that returned the remote type.  Just extend
> it and override that method.  That would be a hack,
> but it's easy since JCS is pluggable.
> 
> --- Aaron Smuts <as...@yahoo.com> wrote:
> 
> > The remote cache doesn't work the way you described
> > exactly.  If a remove request goes to the remote
> > server, it sends it to all listeners and to all
> > auxiliaries.  There is no command to remove from
> > memory only.  
> > 
> > Just like the lateral cache, it can be configured to
> > send remove requests.  If you have the lateral
> > client
> > configured to get from the other laterals, then the
> > laterals start to look like a bunch of remote
> > caches. 
> > However, this is not advisable when  you have a buch
> > of machines.  You don't want to have to search 50
> > boxes for every cache miss. . . . .  
> > 
> > If you are using a shared jdbc disk cache, then
> > there
> > is no reason not to use the remote server.  I push
> > over 500 puts a second through a remote cluster and
> > do
> > the same number of gets . . .  If you are worried
> > about memory, then configure the remote server to
> > not
> > store very much or anything in memory.  That's how I
> > use it for most regions.  Use the remote cache as a
> > way to access the JDBC disk cache.  Also, the remote
> > will take care of the invalidation messages for you.
> > 
> > It is the most scallable way to go.  Also, you can
> > partition the data into multiple rmeote clusters in
> > the future if you needed to scale.   . . .
> > 
> > Hmmn.  I see what you are trying to do.  If the disk
> > cache was configured as a non-local auxiliary (i.e.
> > remote or lateral) then yes, the remove would not
> > remove the item from disk.  What we need is a new
> > type
> > called shared or remote disk.  This would do the
> > trick.        The assumption that all disk caches
> > are
> > local is false in this case.  Perhaps we could use
> > the
> > disk mode setting.   . . . Let me think about it a
> > bit
> > more.  For now, I suggest that you use the remote
> > cache.
> > 
> > Cheers,
> > 
> > Aaron
> >  
> > 
> > --- Randy Watler <wa...@wispertel.net> wrote:
> > 
> > > We are attempting to use a JDBCDiskCache in a
> > shared
> > > mode with multiple
> > > cache clients using it as their backing store. We
> > > are also interested in
> > > using the TCPLateralCache in "remove on put" mode
> > to
> > > help keep the
> > > clients roughly in sync where their extents
> > overlap,
> > > (albeit not at a
> > > transactional level). We are not using a remote
> > > cache because each cache
> > > client will have a largely orthogonal working set
> > in
> > > relation to the
> > > other clients and that would force the single
> > remote
> > > cache to hold a
> > > very large working set for no advantage.
> > > Conceptually speaking, we are
> > > attempting to use the JDBCDiskCache as a remote
> > > cache instead.
> > > 
> > > Therein lies a problem though. Because the
> > > JDBCDiskCache is in fact a
> > > disk cache, JCS propagates lateral updates as
> > > removes to this auxiliary
> > > cache. This ends up actually removing the elements
> > > from the
> > > JDBCDiskCache instead of simply removing the items
> > > from the memory cache
> > > as would happen if it were a remote cache.
> > > 
> > > I am wondering if it would be reasonable to allow
> > a
> > > JDBCDiskCache to
> > > masquerade as a remote cache by indicating that
> > its
> > > cache type is
> > > remote? This would seemingly allow the cache to
> > > operate as a "remote"
> > > disk cache. True?
> > > 
> > > Any comments/tips would be appreciated. Thanks,
> > > 
> > > Randy
> > > 
> > > 
> > > 
> > >
> >
> ---------------------------------------------------------------------
> > > To unsubscribe, e-mail:
> > > jcs-users-unsubscribe@jakarta.apache.org
> > > For additional commands, e-mail:
> > > jcs-users-help@jakarta.apache.org
> > > 
> > > 
> > 
> > 
> >
> ---------------------------------------------------------------------
> > To unsubscribe, e-mail:
> > jcs-users-unsubscribe@jakarta.apache.org
> > For additional commands, e-mail:
> > jcs-users-help@jakarta.apache.org
> > 
> > 
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jcs-users-help@jakarta.apache.org
> 
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jcs-users-help@jakarta.apache.org


Re: JDBCDiskCache as Remote Cache

Posted by Aaron Smuts <as...@yahoo.com>.
You could of course, create a wrapper for the jdbc
disk cache that returned the remote type.  Just extend
it and override that method.  That would be a hack,
but it's easy since JCS is pluggable.

--- Aaron Smuts <as...@yahoo.com> wrote:

> The remote cache doesn't work the way you described
> exactly.  If a remove request goes to the remote
> server, it sends it to all listeners and to all
> auxiliaries.  There is no command to remove from
> memory only.  
> 
> Just like the lateral cache, it can be configured to
> send remove requests.  If you have the lateral
> client
> configured to get from the other laterals, then the
> laterals start to look like a bunch of remote
> caches. 
> However, this is not advisable when  you have a buch
> of machines.  You don't want to have to search 50
> boxes for every cache miss. . . . .  
> 
> If you are using a shared jdbc disk cache, then
> there
> is no reason not to use the remote server.  I push
> over 500 puts a second through a remote cluster and
> do
> the same number of gets . . .  If you are worried
> about memory, then configure the remote server to
> not
> store very much or anything in memory.  That's how I
> use it for most regions.  Use the remote cache as a
> way to access the JDBC disk cache.  Also, the remote
> will take care of the invalidation messages for you.
> 
> It is the most scallable way to go.  Also, you can
> partition the data into multiple rmeote clusters in
> the future if you needed to scale.   . . .
> 
> Hmmn.  I see what you are trying to do.  If the disk
> cache was configured as a non-local auxiliary (i.e.
> remote or lateral) then yes, the remove would not
> remove the item from disk.  What we need is a new
> type
> called shared or remote disk.  This would do the
> trick.        The assumption that all disk caches
> are
> local is false in this case.  Perhaps we could use
> the
> disk mode setting.   . . . Let me think about it a
> bit
> more.  For now, I suggest that you use the remote
> cache.
> 
> Cheers,
> 
> Aaron
>  
> 
> --- Randy Watler <wa...@wispertel.net> wrote:
> 
> > We are attempting to use a JDBCDiskCache in a
> shared
> > mode with multiple
> > cache clients using it as their backing store. We
> > are also interested in
> > using the TCPLateralCache in "remove on put" mode
> to
> > help keep the
> > clients roughly in sync where their extents
> overlap,
> > (albeit not at a
> > transactional level). We are not using a remote
> > cache because each cache
> > client will have a largely orthogonal working set
> in
> > relation to the
> > other clients and that would force the single
> remote
> > cache to hold a
> > very large working set for no advantage.
> > Conceptually speaking, we are
> > attempting to use the JDBCDiskCache as a remote
> > cache instead.
> > 
> > Therein lies a problem though. Because the
> > JDBCDiskCache is in fact a
> > disk cache, JCS propagates lateral updates as
> > removes to this auxiliary
> > cache. This ends up actually removing the elements
> > from the
> > JDBCDiskCache instead of simply removing the items
> > from the memory cache
> > as would happen if it were a remote cache.
> > 
> > I am wondering if it would be reasonable to allow
> a
> > JDBCDiskCache to
> > masquerade as a remote cache by indicating that
> its
> > cache type is
> > remote? This would seemingly allow the cache to
> > operate as a "remote"
> > disk cache. True?
> > 
> > Any comments/tips would be appreciated. Thanks,
> > 
> > Randy
> > 
> > 
> > 
> >
>
---------------------------------------------------------------------
> > To unsubscribe, e-mail:
> > jcs-users-unsubscribe@jakarta.apache.org
> > For additional commands, e-mail:
> > jcs-users-help@jakarta.apache.org
> > 
> > 
> 
> 
>
---------------------------------------------------------------------
> To unsubscribe, e-mail:
> jcs-users-unsubscribe@jakarta.apache.org
> For additional commands, e-mail:
> jcs-users-help@jakarta.apache.org
> 
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jcs-users-help@jakarta.apache.org


Re: JDBCDiskCache as Remote Cache

Posted by Aaron Smuts <as...@yahoo.com>.
The remote cache doesn't work the way you described
exactly.  If a remove request goes to the remote
server, it sends it to all listeners and to all
auxiliaries.  There is no command to remove from
memory only.  

Just like the lateral cache, it can be configured to
send remove requests.  If you have the lateral client
configured to get from the other laterals, then the
laterals start to look like a bunch of remote caches. 
However, this is not advisable when  you have a buch
of machines.  You don't want to have to search 50
boxes for every cache miss. . . . .  

If you are using a shared jdbc disk cache, then there
is no reason not to use the remote server.  I push
over 500 puts a second through a remote cluster and do
the same number of gets . . .  If you are worried
about memory, then configure the remote server to not
store very much or anything in memory.  That's how I
use it for most regions.  Use the remote cache as a
way to access the JDBC disk cache.  Also, the remote
will take care of the invalidation messages for you. 
It is the most scallable way to go.  Also, you can
partition the data into multiple rmeote clusters in
the future if you needed to scale.   . . .

Hmmn.  I see what you are trying to do.  If the disk
cache was configured as a non-local auxiliary (i.e.
remote or lateral) then yes, the remove would not
remove the item from disk.  What we need is a new type
called shared or remote disk.  This would do the
trick.        The assumption that all disk caches are
local is false in this case.  Perhaps we could use the
disk mode setting.   . . . Let me think about it a bit
more.  For now, I suggest that you use the remote
cache.

Cheers,

Aaron
 

--- Randy Watler <wa...@wispertel.net> wrote:

> We are attempting to use a JDBCDiskCache in a shared
> mode with multiple
> cache clients using it as their backing store. We
> are also interested in
> using the TCPLateralCache in "remove on put" mode to
> help keep the
> clients roughly in sync where their extents overlap,
> (albeit not at a
> transactional level). We are not using a remote
> cache because each cache
> client will have a largely orthogonal working set in
> relation to the
> other clients and that would force the single remote
> cache to hold a
> very large working set for no advantage.
> Conceptually speaking, we are
> attempting to use the JDBCDiskCache as a remote
> cache instead.
> 
> Therein lies a problem though. Because the
> JDBCDiskCache is in fact a
> disk cache, JCS propagates lateral updates as
> removes to this auxiliary
> cache. This ends up actually removing the elements
> from the
> JDBCDiskCache instead of simply removing the items
> from the memory cache
> as would happen if it were a remote cache.
> 
> I am wondering if it would be reasonable to allow a
> JDBCDiskCache to
> masquerade as a remote cache by indicating that its
> cache type is
> remote? This would seemingly allow the cache to
> operate as a "remote"
> disk cache. True?
> 
> Any comments/tips would be appreciated. Thanks,
> 
> Randy
> 
> 
> 
>
---------------------------------------------------------------------
> To unsubscribe, e-mail:
> jcs-users-unsubscribe@jakarta.apache.org
> For additional commands, e-mail:
> jcs-users-help@jakarta.apache.org
> 
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: jcs-users-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jcs-users-help@jakarta.apache.org