> actually, i was asking about caching [saving] the same
> object on more than once cache... if u were using the
> Alteon, how would u ensure that aside from getting
> load balancing advantages, u don't end up caching the
> same object on all your servers...?..
The objects will be duplicated no matter how you config squid, because
squid will go direct when refresh. Unless the correct request is being
send to the correct squid server, either by using load balancer or another
layer of squid.
> the idea, i think, would be to cache an object once,
> on any of the servers, and if a user requests an
> object, the squid servers would probably use ICP to
> check with each other whether the object is cached,
> and if not, retrieve it.. but then the other squid
> servers would have to know that one of them has the
> object, and therefore, shouldn't cache it...
>
> if this were case, which would be a very efficient way
> of utilising resources, wouldn't this present a
> latency issue.. because as a user waits for his
> browser to respond to a search, squid servers are busy
> talking to each other at the back end... please
> advise...
Agree. That's why is a balancing between hit rate, latency and high
availability. Using ICP and cache digest will generate background traffic
and higher latency. The extreme case will be to redirect based on url (eg.
based on the first character or url hashing). But, the setback is if one
server down, objects will be redistributed and cause duplication...
Since more than 50% of the internet traffic goes to .com, we figure that
it might be worth the effort to try separating .com and !.com to improve
the hit rate.
Rgds,
Wei Keong
On Thu, 18 Jul 2002, [iso-8859-1] Mark Tinka wrote:
> actually, i was asking about caching [saving] the same
> object on more than once cache... if u were using the
> Alteon, how would u ensure that aside from getting
> load balancing advantages, u don't end up caching the
> same object on all your servers...?..
>
> the idea, i think, would be to cache an object once,
> on any of the servers, and if a user requests an
> object, the squid servers would probably use ICP to
> check with each other whether the object is cached,
> and if not, retrieve it.. but then the other squid
> servers would have to know that one of them has the
> object, and therefore, shouldn't cache it...
>
> if this were case, which would be a very efficient way
> of utilising resources, wouldn't this present a
> latency issue.. because as a user waits for his
> browser to respond to a search, squid servers are busy
> talking to each other at the back end... please
> advise...
>
> AKNIT
>
> --- Wei Keong <chooweikeong@pacific.net.sg> wrote: >
> I suppose you are referring to having duplicate
> > objects in the layer 1
> > squid servers? Is fine actually, since the layer 1
> > squid dont cache any
> > objects, just forward requests and replies.
> >
> > Our concern is to achieve higher hit rate in the
> > layer 2 squid servers.
> > If it is beneficial, we can further separate the
> > layer 2 squid servers to
> > smaller clusters to handle specific domain range.
> >
> > Rgds,
> > Wei Keong
> >
> >
> > On Wed, 17 Jul 2002, [iso-8859-1] Mark Tinka wrote:
> >
> > > just a thought i've developed... when u are using
> > the
> > > Alteon load balancer, how do u ensure that u don't
> > > have duplicate information across your cache
> > > servers.... do u do this by using Squid
> > cache-peering
> > > feature... this is important, as u wouldn't want
> > to
> > > have the same object on each cache server... as u
> > > maximise efficiency and hit rate...
> > >
> > > AKNIT
> > >
> > >
> > > --- Wei Keong <chooweikeong@pacific.net.sg>
> > wrote: >
> > > Hi,
> > > >
> > > > After trying for the past few days, seems like
> > I've
> > > > no luck with the
> > > > Alteon domain redirection/load balancing.
> > > >
> > > > Our current implementation is such that all
> > Squid
> > > > boxes (>10) are load
> > > > balanced by the Alteon. We are thinking of using
> > > > Squid to forward requests
> > > > to separate Squid pool based on domain.
> > > >
> > > >
> > > > | Alteon |
> > > > / \
> > > > |Squid| |Squid| layer 1
> > > > | \ / |
> > > > | \ / |
> > > > | / \ |
> > > > | / \ |
> > > > |Squid| |Squid| layer 2
> > > > .com !.com
> > > >
> > > >
> > > > Essentially, the first layer Squid will only
> > examine
> > > > requests and forward
> > > > to second layer Squid. However, we are yet to
> > test
> > > > the load and
> > > > latency of deploying this architecture. And,
> > > > ultimately, we need to
> > > > find out whether this will bring significant
> > > > benefits (hit rate).
> > > >
> > > > So, squid guru, any input on such an
> > implementation?
> > > >
> > > >
> > > > Rgds,
> > > > Wei Keong
> > > >
> > > >
> > > >
> > > > On Mon, 15 Jul 2002, Wei Keong wrote:
> > > >
> > > > > Ya, have done testing based on the examples in
> > the
> > > > documentation. But, it
> > > > > doesn't seem to work.
> > > > >
> > > > > Understand that it will not affect Squid in
> > > > anyway, just want to see if
> > > > > anyone has more info on Alteon.
> > > > >
> > > > > I am also looking into using CARP to do url
> > load
> > > > balancing/redirection
> > > > > (eg. separate .com and !.com). However, I am
> > not
> > > > sure if it will be bottle
> > > > > neck for request processing. Do you know of
> > > > successful implementation of
> > > > > large-scale CARP? Besides, L4-7 switch and
> > CARP,
> > > > is there any way to do
> > > > > url load balancing/redirection?
> > > > >
> > > > > Rgds,
> > > > > Wei Keong
> > > > >
> > > > >
> > > > > On Sun, 14 Jul 2002, Henrik Nordstrom wrote:
> > > > >
> > > > > > The Alteon documentation should contain
> > quite
> > > > detailed descriptions of
> > > > > > how it works and how to set it up.
> > > > > >
> > > > > > Note: This does not really affect Squid in
> > any
> > > > manner. Is just a
> > > > > > method to get the requests to the correct
> > Squid
> > > > server.
> > > > > >
> > > > > > Regards
> > > > > > Henrik
> > > > > >
> > > > > >
> > > > > > On Sunday 14 July 2002 17.02, Wei Keong
> > wrote:
> > > > > > > Hi,
> > > > > > >
> > > > > > > I am working on url load balancing squid
> > boxes
> > > > using Alteon
> > > > > > > AceDirector. Unfortunately, there isn't
> > much
> > > > resources/docs
> > > > > > > available.
> > > > > > >
> > > > > > > Has anyone tried Alteon url load balancing
> > > > before? Is there any
> > > > > > > forum/discussion group on this? Where can
> > i
> > > > find more info?
> > > > > > >
> > > > > > > Thanks,
> > > > > > > Wei Keong
> > > > > >
> > > > >
> > > > >
> > > >
> > >
> > > __________________________________________________
> > > Do You Yahoo!?
> > > Everything you'll ever need on one web page
> > > from News and Sport to Email and Music Charts
> > > http://uk.my.yahoo.com
> > >
> >
>
> __________________________________________________
> Do You Yahoo!?
> Everything you'll ever need on one web page
> from News and Sport to Email and Music Charts
> http://uk.my.yahoo.com
>
Received on Thu Jul 18 2002 - 01:24:30 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:09:17 MST