RS/6000-C10 Postscript performance driving a 3827
ITEM: RTA000090427
Q:
ABSTRACT: RS/6000-C10 Postscript performance driving a 3827
without AFIG and DPE.
SEARCH ARG: psf/6000 performance
TOPIC THREAD: PRINT
AFP
PSF6000
..
We are demonstrating a proof of concept to a client who has a 3827
without AFIG and DPE to print Postscript files. We have reviewed the
performance documents recently released on the PSF/6000 which discussed
rated speed performance on a 3829 with AFIG & DPE on a 370.
We will put up C10 running PSF/6000 to drive the 3827. What is the
expected performance for this configuration without the AFIG and DPE
features? We understand the IM1 is much more CPU intensive than the
IO1 raster patterns.
A:
I checked with the performance analyst. The printer without AFIG or DPE
will definitely be the bottleneck. If the PostScript transform outputs
a full page IM1 image from a PostScript text file, the printer will
probably only print about 30 IPM. The PostScript transform in PSF/6000
does only produce a rectangular image that is large enough to contain
all the black pels, so the actual IM1 image may not be full page. If,
for example, the result of the transform were 50 square inches of IM1
image, then we would expect the 3827 to print around 52 IPM.
We feel that the Model C10 should be powerful enough to support the
3827 at 52 IPM. If the customer were to upgrade the printer with
AFIG/DPE or to purchase an AFCCU printer that can handle compressed
image so that the printer is no longer the bottleneck, then we would
need to know more about the complexity of the data stream before saying
that the C10 will be sufficiently powerful.
I hope that helps.
Q:
OK, we are going to position to client that without AFIG & DPE, you can
expect less performance than with. We will have available the cards for
testing if they agree to it with the proviso that they will order the
cards if they improve performance.
The page contents are manuals very much like our documentation with
the image content being examples of the screen that one can see when
they are using whatever software the manual is about. These pages are
probably most like PSTXCLIM, other pages are more like PSTXCBAS. I
would probably say that there are more of the pages without images than
with. Hope this can help determine expected performance.
We would like to have soem expected numbers from the above info with
the following:
- C10 without AFIG & DPE
- C10 with AFIG & DPE
- What size RS/6000 with AFIG & DPE to run at 92 IPM?
A:
PSTXCLIM is just about full page, so let's assume that it produces a
full page image; as mentioned above, the expected throughput on the
3827 without AFIG/DPE would be 29-30 IPM. Since your customer has this
configuration today, you can observe actual throughput rather than
relying on our performance estimates in this particular case. We
estimate that PSTXCBAS would print approximately 25% faster, so you
might see 40 IPM on the 3827 without AFIG/DPE. (This information based
on hardware performance numbers in FLASH 9242.)
If you add AFIG/DPE to the printer, performance numbers from FLASH
9242 indicate that a full page compressed image (RAST87) would be able
to print at rated speed (or 92 IPM) on the 3827, assuming no CPU
bottleneck. One caveat is that this is not full page PostScript image
(that is, image imbedded in the PostScript file), but rather a full page
of IO1 image produced out of the PostScript transform--a subtle, but
critical distinction.
Assuming the printer can then print at rated speed assuming no CPU
bottleneck, you want to know:
a) At what rate can the C10 drive the 3827 with AFIG/DPE, assuming
compressed image (IO1_G4) as output? Is the C10 a bottleneck?
b) What model RS/6000 will be necessary to drive the 3827 with AFIG/DPE,
assuming compressed image (IO1_G4) as output, if the C10 can't?
c) And, from offline conversations, I believe the customer actually
has two 3827s. What does that imply for the CPU?
Keeping in mind that we do not have empirical data that map to your
particular configuration, let me provide rough estimates based on the
data that we do have.
a) We estimate that for sample job PSTXCLIM, it would require virtually
100% of a Model C10 to transform and print 92 IPM. We base this
estimate on the performance of that sample job on a Model 370 (using
processor seconds/page, as shown in the methodology in the PSF/6000
Performance Guide on pages 32ff, adjusted for a Model C10 according
to the published SPECint92 and SPECfp92 benchmarks.
Therefore, it is likely that the C10 is not sufficiently powerful
to transform and print this sample PostScript job at a rate of 92
IPM.
b) Since the C10 is not sufficiently powerful, we repeated these
calculations adjusting for a Model C20 processor. We estimate that
transforming and printing PSTXCLIM at 92 IPM would consume
approximately 55-60% of a Model C20 processor. Similar calculations
for a 390H show an estimated utilization of approximately 50%.
c) Two 3827s equate to a desired printing rate of 92*2 or 184 IPM.
Based on the calculations shown in Step b, if your jobs are similar
to PSTXCLIM, we recommend a separate processor for each 3827.
Alternatively, you could separate the transform and printing
processes onto multiple CPUs, placing the transform process on the
more powerful of the CPUs.
In your phonemail to me, you said something about RAST24 as a case.
Were you referring to a PostScript page (no PS image) that would rip
to a page size of 24 square inches of image, or did you actually mean
that the PostScript job contains a PS image block of 24 square inches?
It makes a BIG difference¢ If the latter, you'd be consuming 100%
of a Model C20 to transform and print 92 IPM. If the former, you're
talking about a case of less IO1 image than PSTXCLIM -- about 70% less
I expect. So the type of PostScript to be printed is critical.
I hope that helps provide additional guidance. Since you have AFIG
and DPE cards you can loan this customer, you might want to go ahead
and try them. Then you can see for yourself how fast the C10 can
rip and print the customer's job. We'd certainly be interested in
hearing the results¢
S e a r c h - k e y w o r d s:
AIX PSF/6000 PSF FOR AIX PERFORMANCE AFIG IMAGE 3827 POSTSCRIPT IM1
WWQA: ITEM: RTA000090427 ITEM: RTA000090427
Dated: 07/1999 Category: XPSF6000
This HTML file was generated 2000/11/30~13:34:04
Comments or suggestions?
Contact us