Help with PSF/6000 print server solution and sizing.
ITEM: RTA000091213
Q:
ABSTRACT: Help with PSF/6000 print server solution and sizing.
SEARCH ARG: psf/6000
TOPIC THREAD: PRINT
AFP
PSF6000
..
I have a customer that has an MVS/CICS system as well as a VSE system
and a VM system. They want to be able to consolidate their printing
and have a print server that will allow printing from multiple systems
on the same printers. Right now they have about 100 PCL printers at-
tached on Novell LAN. I'd like to run by you a solution we'd like to
put in front of them.
From VSE and VM we'd funnel the print jobs thru JES. On MVS/CICS we'd
use NETSPOOL to get output into JES. We'd then use MVS Download to
get output into PSF/6000. The RS/6000 would take the linedata and con-
vert it to IPDS for our printers using PSF/6000, or transform it to
AFPDS and PCL for the HP printers.
Does this sound like the best way to go?
Part II:
If this does sound like the best way to go, I'd like to get some siz-
ing help. Item RTA000090012 sounds very similar to what I want to do.
With the following changes to this item, would it be possible to get
an idea of a config needed to support this on the RS/6000 side? We'd
really like to show the customer that he'd be much smarter to put in
3112 type devices to allow PCL printing via their novell printer and
do IPDS printing from the host.
The print file would be line text.
Jobs would average 10 pages each.
Simple text output (NO images, logos, overlays, etc) - 8K per page
Applications that produce output would be CICS, JES, PROFS, stat reports
Printers would be 12 ipm, PCL printers (HPs)
Printers would be attached via SPX/IPX (novell)
100 Printers on novell lan
Standalone system(this will be new system to do this)
Plan on 25% of printers to be active at same time.
I know that using IPDS printers to do this will take considerably less
machine to accomplish and get better thruput, but how much?
I've scanned the PSF/6000 performance information, but I'm having major
problems understanding it. It also makes comments that this data is
only good for 1.1, not 1.2, and that the numbers don't include all
the time needed on the PCL side.
Please help. Thanks.
A:
Commenting only on the movement of host data to the JES spool,
yes, I would transfer the VM and VSE output to the JES spool,
and use NetSpool to capture the CICS output and place it on
the JES spool. Assuming they have/want TCP/IP, MVS Download
would be a good way to get it to the RS/6000. Our PSF/6000
specialist will respond with sizing information.
A:
I've done an estimated sizing for your environment, assuming 25 12-IPM
printers on average active at a time, printing an 8K page with a pagedef
(through ACIF/line2afp). I've sized it once assuming IPDS printers,
and once assuming PCL printers.
Caveats:
1) The data we have for driving PCL printers is based on PSF/6000 V1.1
as you noted. There have been some performance improvements made to
this area of the code, though I don't have quantifiable numbers. But
performance under V1.2 or V2.1 should not be any poorer than under
V1.1.
2) The data we have for driving PCL printers was collected with the
printers directly parallel-attached to the RS/6000. The then-current
versions of the AIX parallel port driver appeared to take quite a bit
of CPU overhead. So a newer version of the AIX parallel port driver
could result in improved performance and less CPU overhead. Also,
if the PCL printer is network-attached, we believe that the CPU
would be less, but we have no quantifiable numbers to tell us how
much less. My AIX printer contacts have no data in this area
either unfortunately. Nor can they tell me how much memory you
should install to drive this many AIX printers. I have a call into
another group to see if they have any information, and if so, I will
update this item.
3) We have not done any testing with routing PSF/6000 or PSF/AIX output
to Novell-attached printers. We suspect our estimates of AIX CPU
utilization will be high given the same considerations in caveat #2.
Note that with FLeX/IP on the Novell server, jobs will spool again
at the Novell server so you'll have triple-spooling (PSF, AIX, and
Novell).
Having said all that, let me run through my calculations. I'm using
the methodology described at the back of the PSF/6000 Performance
document beginning on p32.
25 active printers * 12 ipm = 300 ipm that PSF/AIX must produce in order
to have sufficient data to feed the printers at rated speed.
1) Assuming IPDS printers:
Using Table 3 in the PSF/6000 Performance Information document dated
5/18/95, assuming a job complexity similar to LTX1U08K, you can see
that the processor seconds per page (rightmost columns) for an RS/6000
Model 370 come out to .025 for the ACIF transform piece and .018 for
the printing piece, for a total of .043 M370 processor seconds per
impression.
.043 M370 CPUsecs/impression * 300 IPM = 12.9 M370 secs/min
12.9 M370 secs/min / 60 secs/min = .215 seconds of CPU time for every
second of printing at that rate.
That equates approximately to 21.5%
of a Model 370 for this workload.
So a Model 370 is overkill. How does a Model 250 compare? A M370
is 1.1-1.6 times as powerful as a Model 250 according to the SPECint92
and SPECfp92 benchmarks, so I would expect the load on a Model 250 to
be approximately anywhere between 24-35% of a Model 250, still plenty
of room. You might also consider the 43P--about the same base price for
the system unit as the 250, but with a whole lot more power˘ I've not
done a comparison of a fully configured 250 vs 43P, but would like to.
One note: the 43P does not support the S/370 Channel Emulator/A adapter
as it has no microchannel slots. But in your case, that doesn't matter.
2) Assuming PCL printers:
Using Table 11 in the PSF/6000 Performance Information document, with
the data for a PCL5 printer and a job similar to TXT08K, in addition to
the ACIF transform data from Table 3 (since the file must still be run
through ACIF if mapping with a pagedef), you have .025 M370 CPU seconds
for the ACIF transform piece (Table 3) and 0.260 M370 CPU seconds for
the subsequent conversion to PCL and printing (Table 11). (The caveats
I mentioned above relate to the 0.260 number; we believe it's high but
don't know how much.)
.025 + .260 = .285 M370 CPUsecs/impression
.285 M370 CPUsecs/impression * 300 IPM = 85.5 M370 secs/min
85.5 M370 secs/min / 60 secs/min = 1.43 seconds of CPU time for every
second of printing at that rate or
approximately 143% of a Model 370.
Thus, a Model 370 would be
insufficient for this workload if
our data/estimates are adequate.
You would either need to consider two RS/6000s to drive this, or
perhaps a single, more powerful RS/6000 such as the 43P. The 43P
(7248-133) is 1.3-2.5 times more powerful than the 370 according to
the SPEC benchmarks I've seen, and less expensive to boot˘ If you
assume conservatively that the 43P is only 1.3 times more powerful
than the 370, then the 43P is still not powerful enough to drive
25 12IPM PCL printers simultaneously using the data that we have.
So you can see there's a BIG difference in driving IPDS printers vs
PCL printers from a standpoint of CPU utilization alone. There are
also differences in the amount of data going over the network in most
cases, as the PCL output can be quite a bit larger than the IPDS data.
And this will also impact the amount of fixed disk needed for spool
space on the RS/6000 as well. The cost justification between using a
less expensive RS/6000 to drive IPDS printers may offset the additional
cost of the IPDS printers and the i-data 7913˘
I don't have any direct comparisons of PCL vs IPDS, but we do have some
customer experience comparing PCL vs AFPDS (which should be pretty darn
close). I'll include the customer information below, but keep in mind
that it is their experience only with their apps, and that other
customers should consider conducting their own tests for the purposes
of comparison. I hope this helps. Please reopen if you need additional
information.
***********************************************************************
A new PSF/6000 customer noticed that in his environment, AFP produces
significantly smaller printer data streams than Postscript or PCL.
They were very interested in this due to the efficiency of AFP relative
to sending print to several thousand printers on a World Wide Network.
Further examination is showing that the experience at this customer is
representative of the AFP data stream when compared to PCL or
Postscript.
They tested 11 sample applications from Windows. In this group the
AFP driver created a printer data stream that averaged 1/2 the size of
the same output driven by PCL or Postscript. Thus the resources needed
to transmit and store AFP would be 1/2 of what would be needed for
PCL or Postscript. As the detail below indicates, the data stream sizes
vary, and AFP is not smaller for every application. However, overall:
Postscript was 2.37 x the size of AFP
HPPCL was 2.04 x the size of AFP
The printer data streams were created using Microsoft Windows 3.1 and
the following printer drivers:
AFP Windows driver AFP
HP 3si HPPCL driver PCL
HP 3si Postscript Postscript
PRINTER DATA STREAM
SIZE IN 1000 BYTES
Application Description PS PCL AFP
------------------------------ ----- ----- -----
1 EXCEL 07.XLS 35.3 271.9 293.6
2 ALLIMAGE.DOC 645.8 692.8 561.2
3 DLNEWS 297.5 1267.8 206.5
4 ALLIMAGE.WP 3390.2 264.2 153.3
5 DISEDIT.DOC 1170.8 919.1 832.2
6 EXCEL.DOC 20.1 86.1 123.0
7 WORD DOCUMENT 127.1 256.5 105.4
8 READ FIRST MS 3.1 (28 PAGES) 130.0 120.1 112.3
9 WINDOW WORKS LETTER (1 PAGE) 12.1 6.7 1.1
10 POWER POINT (48 PAGES) 278.0 1370.0 188.0
11 AMIPRO LETTER 40.5 13.1 8.1
------ ------ ------
TOTAL BY PRINTER DATA STREAM 6147.4 5268.3 2584.7
SIZED COMPARED TO AFP (X) 2.37 2.04 1
***********************************************************************
S e a r c h - k e y w o r d s:
AIX PSF/6000 PSF FOR AIX PERFORMANCE PCL IPDS AFP NOVELL
psf/aix sizing configuration rs/6000 comparison compare
WWQA: ITEM: RTA000091213 ITEM: RTA000091213
Dated: 05/1999 Category: XPSF6000
This HTML file was generated 2000/11/30~13:34:04
Comments or suggestions?
Contact us