Re: AW: [Handle-info] CNRI Handle Extension for Firefox

[Sean Reilly <sreilly@cnri.reston.va.us>]

Hi John,

Thanks for your message.

While following HTTP links is currently how most end-users view  
handles and DOIs, there are many other Handle/DOI resolutions that  
don't involve web browsers and require the scalability built into the  
"native" resolution.  The benefit of the firefox extension is that it  
provides the needed scalability while being compatible with the http://proxy/identifier 
 style of reference and therefore invisible to end users.  While that  
invisible benefit isn't a great marketing ploy, we are working on ways  
of making the extension more appealing in more visible ways.

A few reasons to have a native handle client are scalability,  
reliability, security and performance.

Performance, with the obvious exception of those behind strict  
firewalls, will be far better (maybe even noticeably!) for clients  
that are capable of sending/receiving UDP messages.  While a native  
client can resolve a handle in under 50 ms it will take a browser  
about ten times that long to resolve hdl.handle.net, setup a TCP  
connection, send the handle/http request to a proxy server, wait for  
the response, receive the response and break down the connection.

Security is always improved using the native resolution protocol  
because it avoids the possibility of DNS cache poisoning, DNS messages  
spoofing, and general interference with the non-SSL hdl.handle.net or  
dx.doi.org resolutions.  A native client can always enable signed  
handle resolution which will only scale when using the native  
protocol.  SSL/TLS-enabling hdl.handle.net and dx.doi.org is not a  
practical option.

Reliability... the HTTP protocol has no built-in reliability, making  
the proxy (or any other HTTP service) subject to more outages than  
should be necessary.  If a web browser resolves a domain name to 10  
different IP addresses, the browser (or OS) picks one and then tries  
to contact it.  If that contact fails, then the browser gives up.   
Native handle resolution, like DNS, will try the next server in the  
list until one of them responds.  This built-in reliability means that  
servers that go down don't have to mean a loss of service for  
clients.  We go through great effort to make hdl.handle.net and  
dx.doi.org reliable and accessible 24/7/365 however you can only get  
so far with a system that doesn't have reliability baked in from the  
beginning.

As for scalability, as the number of DOIs and handles continue  
increasing, the proxy system will eventually reach a point where it  
receives too much traffic and is too costly to maintain.  That is,  
unless at least some of the load can be moved to the distributed  
native protocol.  The Handle System is a distributed name resolution  
system, designed to be more scalable while allowing a flatter  
namespace than DNS.  Dependence upon the proxies throws out that  
advantage.  Consider that as of around 2004 DNS had about 38 million  
domains under the com/net/org/info/biz TLDs.  If everyone tried to  
resolve DNS names via an HTTP service, that HTTP service, no matter  
how tricked out with load balancers and server clusters would be  
crushed under the load.


Cheers,
Sean


On Mar 11, 2009, at 8:49 AM, John S. Erickson wrote:

> The following was asked earlier, but in gentler terms: Could someone  
> please explain to me the point of the native service, if the  
> rollover and default answer in (arguably) the common case  
> (organizations behind firewalls) is http and the proxy?
>
> Given the performance numbers cited in this thread (and if I  
> understand them...), if for some reason I need "native" (quotes  
> intentional) resolution I'll continue to use the iGoogle gadget I  
> wrote a couple years ago (there are now a few others)...which  
> "instantly" resolves and works in all browsers...




_______________________________________________
Handle-Info mailing list
Handle-Info@cnri.reston.va.us
http://www.handle.net/mailman/listinfo/handle-info