Re: DFT, Caching and Riverbed - Oh My!

Mark Hattersley — Wed, 06 Oct 2010 23:45:22 GMT

A few more thoughts -

1. Obviously, Riverbed users need to be "behind the appliance" to benefit from their optimizations - where DFT is available just with the connection between ProjectWise Explorer and Integration Server - for hardware-based accelerators, obviously :-)

2. You can use both - disable dms traffic from Riverbed - let Riverbed optimize Exchange, SAP, web, etc. and let DFT optimize your engineering data

just my 2 cents

mark

Re: DFT, Caching and Riverbed - Oh My!

Dalius Dobravolskas — Tue, 05 Oct 2010 12:32:26 GMT

That's not presentation but I think that might help as well.

There are two things to consider when using DFT: server CPU power and network speed. DFT algorithm is CPU intensive and in local network sending complete file might be faster than DFT. Using real life examples I have seen only one: file operations were 2 times faster without DFT in local network. However I don't know details: what CPU was used, at what time DFT was tested and etc. This example could be very specific to that environment.

While we don't offer DFT load test application there is DFT Benchmark Wizard supplied with standard PW installation. I recommend using this tool on real network during work hours to see if DFT helps you or you should disable DFT in local network. I'm pretty sure that DFT is mandatory for remote locations. DFT can be controlled in dmskrnl.cfg. As I have said better CPU on server might change your test results.

About remote caching servers (officially known as caching gateways). If you have remote team(s) you should seriously consider having caching gateways because of work performance.

Lastly about Riverbed. As John Schown mentioned (here communities.bentley.com/.../140425.aspx) combination of DFT and Riverbed gives only small benefits.

I personally don't know much how Riverbeds works internally while some information is available. Based on information I have found out Riverbed optimizations are done at various levels:

1) MXTCP - this optimization shouldn't have positive or negative effect on DFT. That's general network optimization;

2) HSTCP (en.wikipedia.org/.../HSTCP) - Again does not look like optimization what could make difference for DFT or no DFT. While in some specific (mostly atypical) situations it might give small benefit for noDFT;

3) Riverbed does some local caching. In my opinion DFT data sent over network can't be optimized by Riverbed because different users might have files on their system retrieved at different times. Thus noDFT with Riverbed might give better results than DFT without Riverbed. Evaluation must be done very carefully because that really depends on your work style, data formats and your work environment. E.g. if you make frequent changes that are used by your team I would bet on DFT, but if you make one big change per week Riverbed might be better option. But still that depends on how big are your changes, how big are remote teams and if you are using caching gateways. I can actually put it other way round: depending on your choice not to use DFT you might affect your and your team's work style and performance.