BLD FLASH 9525 SOURCE MATERIAL DATED: 8/95 ¦ PSF/400: CPU CONSIDERATIONS FOR TCP/IP ATTACHMENT As of June 30, 1995, OS/400 V3R1 and V3R6 Print Services Facility/400 will provide direct support for TCP/IP attached printers including 3130,3160 3935 and 3900 AFCCU models and 7913 attached models such as the 3112, 3116, 3912, 3916, 3816, 3930 and 4028. Support for 7913 attachment of the IPDS impact models 4230, 6408/6412, 4224/4234 will also be provided. A complete understanding of the ramifications of moving from SNA or twinax attached printers to TCP/IP attachments should be gained before moving cus- tomers in the direction of TCP/IP attached printer solutions. Although the ability to route print using TCP/IP provides additional printer sharing capabilities and attachment flexibility, the testing in our labs has shown that TCP/IP in the AS/400 environment does not drive printers as effi- ciently as SNA token ring or twinax. For customer environments where CPU cy- cles are in short supply, CPU utilization should be reviewed as part of a system assurance or pre-installtion planning procedure. The performance evaluation figures below should be used only to help under- stand the CPU utilization effects of various attachment methods and not for printer performance comparisons. As of 9/95 there is work in progress both in PSF/400 and TCP/IP for OS/400 development groups to improve the performance and CPU utilization when using TCP/IP. These performance measurements were as of 8/95 and part of your TCP/IP printing with PSF/400 implementation plan should be to contact Software Support to receive the latest maintenance to PSF/400 and TCP/IP for OS/400 to be sure to receive any fixes that may be available. DIRECT TCP/IP ATTACHMENT (SUPPORTED BY AFCCU PRINTERS) ______________________________________________________ The figures shown below were tabulated in the following test environment: AS/400: D60 LAN: Dedicated 16M T.R. MTU: 4096 OS/400: V3R1 Line Max Frame Size: 8156 Traces: Off Print While Convert = *NO PRINTER HARDWARE: 3130 PRINT JOBS: The print jobs used in this performance evaluation are de- scribed in the "AFP: Catalog of Performance Reference Pages". Summary descriptions are also provided below. TXT32K - Simple Text reference page with 32,000 characters. TXT08K - Simple Text reference page with 8,000 characters. RAST87 - Simple Raster reference page with 87 square inches of IM1 image. Total bytes per page, 65,253. Case name Printer ATTACH/TYPE PPM KB/S CPU Util. CPU Sec./Page TXT08K 3130 TCP/IP 30 4.5 0.4% 0.008 TXT08K 3130 SNA 30 4.5 0.2% 0.005 TXT32K 3130 TCP/IP 30 18.1 1.5% 0.030 TXT32K 3130 SNA 30 18.2 1.0% 0.019 RAST87 3130 TCP/IP 17 177.1 12.6% 0.445 RAST87 3130 SNA 21 220.9 8.3% 0.235 NOTES: 1. The 3130GA printer is a ship-level 3130 used at 240-pel density. 2. The 3130M printer was a printer with test microcode required to use TCP/IP send buffers larger than 8K. Contact hardware support, to obtain the applicable microcode upgrade for your specific AFCCU printer. 3. Print File conversion requires CPU seconds and causes utiliza- tions in addition to what's shown for printing, but is not af- fected by attachment types and variables. The total processor load for printing these pages is higher than the printing-only results shown here. If the total CPU Utilization was calcu- lated for Print While Convert= *YES (configured in the Printer Device Description) it is estimated the figures would reflect approximately 20-30% differences between TCP/IP and SNA vs the 50% shown in many of the test cases here. 4. The performance (PPM) degredation shown in the RAST87 test case is believed to be primarily caused by limitations in the AFCCU printer control unit. TCP/IP ATTACH VIA 7913 ______________________ Some TCP/IP 7913-3930 measurements were made, to understand the effects of 7913 limitations and to compare the same device attached via twinax. The print environment was as follows: AS/400: D60 LAN: Dedicated 4M T.R. MTU: 1440 OS/400: V3R1 Line Max Frame Size: 4060 Traces: Off Print While Convert = *NO Print Buffer: 256 bytes * Print buffer used for direct twinax attach was 1024 bytes Case name Printer ATTACH/TYPE PPM KB/S CPU Util. CPU Sec./Page TXT08K 3930 7913 / coax TXT08K 3930 7913/twinax TXT08K 3930 Twinax TXT32K 3930 7913 / coax 15 9.0 1.7% 0.050 TXT32K 3930 7913/twinax 15 9.0 1.7% 0.050 TXT32K 3930 Twinax 16 9.4 .6% 0.023 RAST87 3930 7913 / coax 3 33.3 5.9% 0.827 RAST87 3930 7913/twinax 3 33.3 5.9% 0.799 RAST87 3930 Twinax 3 31.3 2.0% 0.400 NOTES: 1. The larger buffer size used for direct twinax attachment should not affect processor seconds per page, although printer throughput might be affected. 2. The limited throughput in these test cases was primarily caused by the 3930 print engine. 3. The different buffer sizes used for the 7913 attachment is due to a permanent restriction that currently exists in the 7913. These numbers were all gathered using an AS/400 Model D60. Evaluations on your specific model of AS/400 can be made using the IBM AS/400 System Hand- book (GA19-5486). The following factors should be considered when evaluating whether the CPU utilization impact of TCP/IP attachment of AFP printers could affect performance or throughput in your environment. o Relative CPU performance of your specific AS/400 Model. o The number of printers and/or the pages per minute of the printer(s) be- ing driven by PSF/400. o Other applications running on the AS/400 and/or the current CPU utiliza- tion levels. o System Tuning. 3900 DUPLEX TCP/IP TOKEN RING ATTACHED ______________________________________ AS/400: F60 LAN: Dedicated 4M T.R. MTU: 4060 OS/400: V3R1 Line Max Frame Size: 4060 Traces: Off Print While Convert = *NO Case name Printer ATTACH/TYPE PPM KB/S CPU Util. CPU Sec./Page TXT32K 3900 TCP/IP 300 176.8 6.9% 0.014 RAST87 3900 TCP/IP 300 204.4 8.3% 0.254 $EOM