[mapserver-users] Re: [Tilecache] WMS shootout, ESRI and ERDAS

[Bruce Foster]

Richard,

Thanks for responding and also for the explanation.

Well, that was the exact point that I'm trying to establish. For
example, you said that, IWS gets the following numbers;

> 256 x 256 = 420 per second
> 800 x 600 = 92 per sercond

But on a identical configuration, but not exact configuration, the
results that I got were far different from what you have said. OK,
admitting the fact I'm using Jmeter for testing. You may use something
else being a commercial company.

The fact, is the number that I got were far far less. The testing was
done on the bluemarble NG image. The raw image size was 10GB and
compressed to 400mb. RANDOM function of the jmeter was used to
generate totally random bounding boxes. The test was carried out at
one of your customers who wanted to check the performance of the WMS
for a mission. Note, we made sure no second hit goes to the same
bounding box.

So, its quite easy for one to argue the results and I'm sure thats one
of reasons why you dont want to release any numbers, apart from the
competition point of view.

I can release the complete test scenario and the number here for the
public if you allow me to do so. I have no authority to do that
without erdas permission.

It would have been much easier to convince "someone" if we have
numbers from a equal play field. Well, that was your decision and I
respect that. No arguments.

I'm here to answer any question.


---
Bruce
NSW Australia





On Sat, Oct 17, 2009 at 5:49 AM, Richard Orchard
<richard.orchard at erdas.com> wrote:
>
> Hello,
>
> I am the product manager for Image Web Server (http://iws.erdas.com/).  I
> would like to add to the information presented here.
>
> I am not particularly secretive about our performance - although i wouldn't
> necessarily give my "inside" information to a competitor, i am happy to
> share some results with you.  I have the real joy in loving my product - it
> is fantastic!  Although, some people here probably think the love goes a bit
> too far....
>
> As an example:
>
> Intel Xeon E520 2 x Quad Core 2.5 GHz
> 8GB Ram
> RAID 5 Configuration
> 40GB OS partition; 1000GB data partition
> 15,000 RPM SCSI disks  - yes, hard disks make a difference.  Disk IO is a
> big "cost".
> Operating system: Win 2003 R2 SP2 64bit version.
>
>
> Performance numbers as far as WMS requests are concerned are largely
> dependent upon the type (size, compression) of imagery that is being served.
>
> So, taking a small blue Marble image, only about 2.6 GB uncompressed (i used
> an ECW file):
>
> 256 x 256 = 420 per second
> 800 x 600 = 92 per sercond
>
> If we want to up the size of the image, we can take a 2.1 Terabyte landsat
> image of the world.
>
> 256 x 256 = 351 per second
> 800 x 600 = 68 per sercond
>
> So - OK type numbers, and enough to sustain quite a large volume of people.
> A caveat would be that the machine i tested on was our demo machine
> (http://iws.erdas.com/) and i did not "un-plug" it from the world, so other
> requests were no doubt coming in.  It also runs some other applications - so
> it probably could do "more" than this, but i was happy enough at the time,
> so i just left it at that.
>
> If we want to see "rediculous" performance, we can switch to ECWP.  ECWP is
> a high-speed streaming protocol.  ECWP is hard to "benchmark" because i need
> to get so many free client machines.  Anyway, when i have gather XX number
> of client machines - each simulating hundreds of separate clients, we can
> simulate and respond to 5000 or more concurrent users.  At the same time CPU
> usage only gets to ~20%.
>
> For the tile-cache enthusiasts out there, we have an "Optimized Tile
> Delivery" system.  Same sort of "affect" but a different methodology.  I
> wrote a couple of items about it:
>
> http://labs.erdas.com/blog_view.aspx?q=6071
> http://labs.erdas.com/blog_view.aspx?q=6076
>
> The reason why IWS didn't take part in the "shoot-out" was the requirement
> to use a linux operating system.  We focus much more on Windows - although
> our Solaris version is quite well tested.
>
> Now - i would like to say that i am "signing" this information as correct
> and accurate.  you have my word.  But - if you want to try it for yourself,
> and you are genuine, you are able to download IWS from erdas.com and i will
> provide you a license and help you with the installation / setup.
>
>
> Best wishes,
>
>
> Richard Orchard
> http://iws.erdas.com/
>
>
> --
> View this message in context: http://n2.nabble.com/WMS-shootout-ESRI-and-ERDAS-tp3826002p3838528.html
> Sent from the TileCache mailing list archive at Nabble.com.
> _______________________________________________
> Tilecache mailing list
> Tilecache at openlayers.org
> http://openlayers.org/mailman/listinfo/tilecache
>



-- 
Thanks

Bruce