VOExplorer List Directory
Motivation
Simply listing
MySpace directory contents in VOExplorer takes longer than seems reasonable. The aim is to try to isolate the which component is responsible for this slowness. There are several components that are involved in
Methodology
Network traffic from the local host to the various servers was captured using an OS level packet sniffing tool (wireshark) whilst listing the contents of a directly
Results
So from the point where I
press the login button (or at least the first traffic I see as a
result of it) (time in seconds in first column)
| Time(s) | event |
| 0.00 | query galahad for ivo://uk.ac.le.star/ org.globus.myproxy.MyProxy |
| 0.07 | galahad responds "not found" |
| 0.11 | query loki.roe.ac.uk for ivo://uk.ac.le.star/ org.globus.myproxy.MyProxy |
| 0.17 | loki responds with apparently empty soap message?? |
| 0.22 | query galahad for ivo://uk.ac.le.star/ org.astrogrid.community.common.security.service.SecurityService |
| 0.29 | galahad responds with the registry record |
| 0.32 | call corazon.star.le.ac.uk to check the username/password |
| 0.45 | successful login response from corazon.star.le.ac.uk snitch happens |
| 0.60 | query galahad for ivo://uk.ac.le.star/ org.astrogrid.community.common.security.service.SecurityService (again??) |
| 5.66 | galahad reponds. |
| 5.69 | call corazon.star.le.ac.uk to check the username/password |
| 5.81 | corazon responds with success |
| 5.85 | query galahad for ivo://uk.ac.le.star/paulharrison |
| 5.91 | galahad responds "not found" |
| 5.98 | query loki.roe.ac.uk for ivo://uk.ac.le.star/paulharrison |
| 6.04 | loki responds with empty soap message... |
| 6.09 | query galahad for vo://uk.ac.le.star/ org.astrogrid.community.common.policy.manager.PolicyManager |
| 11.15 | galahad responds with registry record... |
| 11.23 | call corazon.star.le.ac.uk with a GetAccount? for ivo://uk.ac.le.star/paulharrison |
| 11.29 | corazon responds with homespace filemanager node. |
| 11.32 | query galahad for ivo://uk.ac.le.star/filemanager |
| 11.40 | galahad responds with record. |
| 11.42 | query makalu.star.le.ac.uk for mypace contents prefetch depth 3 and parents |
| 11.65 | makalu responds. |
There is a shocking finding from the above - it is sometimes taking
the registry about 5 seconds to respond to a "lookup ivorn" query
(which should be the simplest - i.e. do not depend on complex search)
- in fact it seems that "every second query" takes a long time.... -
however, the "fast response" is within 0.2 seconds - clearly if 10 seconds (the two slow registry queries
in the above) were removed from the above sequence the time to
first display of home myspace contents would be relatively snappy 1.5
seconds.
After seeing these result
GaryGilchrist pointed out that galahad.star.le.ac.uk was acting as a proxy for the real registry implementation on rakaposhi.star.le.ac.uk, which has the dns alias of registry.astrogrid.org. The test was rerun with the VOExplorer set directly to point at registry.astrogrid.org and the 5second delays disappeared.
Conclusions
It seems that there was something wrong with the proxying that was occurring on galahad. It would be worth investigating this further to avoid making the same mistake in the future. The main recommendation is, however, to configure clients and services to point to the DNS alias registry.astrogrid.org, which can be adjusted to point directly to the operational registry. Using this alias is better than using a specific machine name, beacause that required a proxy to be installed on the original machine (galahad) when the registry was physically moved in order to avoid having to reconfigure all the services/clients to point to the new registry.
--
PaulHarrison - 11 Jan 2008