Re: XML/A Performance
From: Akshai Mirchandani [MS] (akshaim_at_online.microsoft.com)
Date: 02/12/04
- Next message: Mosha Pasumansky [MS]: "Re: XML/A Performance"
- Previous message: Akshai Mirchandani [MS]: "Re: xslt transforms of large results"
- In reply to: Wayne M: "Re: XML/A Performance"
- Next in thread: Mosha Pasumansky [MS]: "Re: XML/A Performance"
- Messages sorted by: [ date ] [ thread ]
Date: Thu, 12 Feb 2004 12:21:50 -0800
The Integrated Security issue has already been fixed for the 1.1 release.
We'll consider the Nagle algorithm fix a little more carefully before making
the decision...
As far as ADOMD.NET is concerned, you will also have the option of using the
"IXMLA" mechanism to connect to AS2k servers -- i.e. avoid installing XMLA
on a middle-tier and instead install IXMLA and PTS on the client machines.
Then ADOMD.NET will connect through IXMLA+PTS to the AS2k server directly.
This solution will avoid your authentication issue completely.
Thanks,
Akshai
-- This posting is provided "AS IS" with no warranties, and confers no rights Please do not send email directly to this alias. This alias is for newsgroup purposes only. "Wayne M" <wmdev@yahoo.co.uk> wrote in message news:176adae9.0402120735.3bcf4e7f@posting.google.com... > We are using Integrated Security. > > Akshai, your theory seems to support our evidence. If authentication > is being done for every request, then it will be slow. It seems that > the XML/A proxy tries to connect as anonymous first (I think this > confirms to the HTTP 1.1 spec), it gets a HTTP 401 status code and > then it will resend with authentication; this time a successful HTTP > 200 is returned. The evidence for this is in our IIS logs. There are > two posts to msxisapi.dll for every request - the first comes back > with a 401, the second is fine with a 200. > > Both this and Mosha's earlier post regarding Nagling are real concerns > for us. We have a web clustered architecture with a user base of > thousands, and we have engineered our application to be responsive by > issuing relatively small executes, therefore a 200ms latency becomes > more obvious. > > Going forward, ADOMD.NET provides a very nice object model, but if it > is based on XML/A rather than OLEDB then it too will perform badly. > > Are both of these issues being looked at for the XML/A 1.1 release in > a few weeks time? > > Wayne > > "Akshai Mirchandani [MS]" <akshaim@online.microsoft.com> wrote in message news:<OAmUe6O8DHA.3288@TK2MSFTNGP11.phx.gbl>... > > Were you using Integrated Authentication or was it Anonymous? My belief is > > that this perf issue is being caused by the fact that we didn't properly > > support Keep-Alives of HTTP 1.1 and so for authenticated connections the TCP > > connection was being destroyed and re-created for every request. This would > > also cause authentication to be done for every message instead of only once > > for the first connection. > > > > Thanks, > > Akshai
- Next message: Mosha Pasumansky [MS]: "Re: XML/A Performance"
- Previous message: Akshai Mirchandani [MS]: "Re: xslt transforms of large results"
- In reply to: Wayne M: "Re: XML/A Performance"
- Next in thread: Mosha Pasumansky [MS]: "Re: XML/A Performance"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
Loading