Hello
I have been discussing the issues of bandwidth and speed of access with HCN. Although I have not been following the discussion in detail, I have been concerned that some of the discussion has been inaccurate. I have been very blunt with HCN asking them if their servers are slow and causing some speed of access problems in the north west. I have also asked the Managing Director if HCN is 'fobbing off' the north west as one email suggested.
I intend to work via the LIS Patch Groups in those areas were speed is clearly an issue. In the meantime, below is an email from HCN.
Regards
Colin
Please email me direct @ [log in to unmask] <mailto:[log in to unmask]>
COLIN DAVIES.
Deputy Director of Health Libraries - Information Technology.
Northwest Health Care Libraries Unit.
Thelwall House
Lovely Lane
Warrington
WA1 1QG
-----Original Message-----
From: p=NHS NATIONAL INT;a=NHS;c=GB;dda:RFC-822=robin.turner(a)hcn.org.uk;
Sent: 22 October 2001 19:37
To: Davies, Colin
Cc: p=NHS NATIONAL INT;a=NHS;c=GB;dda:RFC-822=allison.hartman(a)hcn.com.au;
Subject: performance issues
Colin
I have looked again at the situation with regard to the apparent speed discrepancies being reported from the North West Region.
Our original starting point was that a significant number of users have always reported a good reponse, Your message seems to confirm this, thus verifying that for them at least, the servers are delivering a good 'turn round', you site 5 seconds as a response below.
What puzzled us was that some individuals reported perfomance results of much longer times, 1 hour. This leads to the conclusion that since all users access the same servers it logically has to be something that effects some users locally. It can't be the servers, or our link bandwidths, or ALL users in ALL areas would be similarly effected, which is not the case.
Also some of those times reported are difficult to accept, the intermediate server 'times out' well within the time limit and no response would be passed. Our original suspicion of interconnect bandwidth does hold true for some local NHSnet connections with bandwidth limitations, but not all. Only those who have a degraded response but probably only within a minute or so.
Investigations recently have led to an interesting development. We believe that in some locations network trafic is cached. This WILL cause a problem since the WebSpirs pages are dynamic and hold session data that is time specific. If a cacheing server intercepts and responds instead of the HCN servers this will interfere with the process and the response will be unpredictable. Our system manager has asked the manchester cachemaster to exclude us from the process which if properly performed, will I firmly believe alleviate the situation.
Interestingly, cacheing is normally in place to improve the response times of networks when there is limited bandwidth which needs to be conserved, and the same pages are being repeatedly accessed. The Manchester cachemasters assertion that there is more than adequate bandwidth seems to contradict the need for cacheing in the first place. A curious paradox.
Also users with a 'stalled' screen get an almost instant response when they hit the F5 key to refresh the screen.
FYI The HCN server configuration is set up so that a dynamic web front end session is run on one set of servers (WebSpirs) to collect the searches which are then passed internally to a database back end (ERL). This back end responds to the internal server over a fast ethernet link which then wraps the answer and presents it back.
The ERL database server is well over specified for the actual load. The response from the Webspirs servers is fairly constant inasmuch that we don't seem to see adverse comments in terms of logging in or displaying the search entry screens, only the database searches.
Internally we test our own service by using a 64k line from Demon to get out to the internet and then back through the bigger link from UUnet to simulate the performance available to customers.
To check, I have just tried a search to Embase resulting in 9994 hits which took 4 seconds to display and 3 to begin to show the results. The more complex search 'calcium and antagonist and unstable and angina' took 6 seconds to search. Not especially slow.
The way to demonstrate is to get the librarians to try it from home via the internet on their own machines and dial up connections.
The comment below in your quote would actually bear out the Cache theory since the download of the PDF file might actually not be coming across the net at all, but from local cache. The user cannot tell ! And having a big connection is a bit like the M6 southbound just north of Birmingham on a wet friday evening, sure it's big - but is it quick ?
Finally, and perhaps most importantly, I would never 'fob off' a customer.
HCN is rightly proud of it's service.
Happy to keep up investigating till we eventually find the bottleneck !!!
Robin Turner
Head of Information Systems
HCN
Charwell House Alton
Hampshire GU34 2PP
Tel: +44 (0) 870 8723 003
Fax: +44 (0) 870 8723 004
E-mail: mailto:[log in to unmask] <mailto:[log in to unmask]>
Web: www.hcn.org.uk <http://www.hcn.org.uk>
'The Leader in Health Internet & Clinical Software'
|