BLD FLASH 9706 SOURCE MATERIAL DATED: 7/97 JUNE 1997 INFOPRINT PERFORMANCE ENHANCEMENTS The IBM Printing Systems Company has significantly enhanced the performance of its InfoPrint Solution for the commercial print on demand market with the availability of the InfoPrint 4000 IR1/IR2 Duplex Printing System and the latest version of the InfoPrint Multiple Printer Controller (RPQs 8B4010 and 8B4011). Both of these products were announced in May 1997 (see announcement letter 197-101) and began shipping in June. Existing InfoPrint customers with RS-6000 39H or F30-based Multiple Printer Controllers (MPCs) and InfoPrint 4000 464/600 (3900 DR1/DR2) Duplex Printing Systems will find significant performance improvements in the following areas: concurrently RIPping and printing files, "saving pages" times (an order of magnitude better in some cases), and performance in "proof" mode (print rates in proof mode with the enhanced products often are about the same as print rates in "saved-page" mode with the prior solution). The purpose of this flash is to describe the enhancements to both the printer and the MPC and discuss how these changes benefit both existing and prospective InfoPrint customers. This flash assumes the reader has already reviewed FLASH 9651 "PERFORMANCE CONSIDERATIONS FOR INFOPRINT SOLUTIONS" and understands such terms as "saving pages", "RIPping files", and "proof mode". The former flash should be available for reference when reviewing the following information. The former flash is available on the PRINTERS disk as FLASH 9651INFP or on HONE in ASKQ using BLD FLASH 9651 as a search argument. As with the prior flash, the information presented here is based on limited performance measurements made in a controlled laboratory environment. InfoPrint Manager V1.3.1.2 (pre-release) was used on the new RS/6000 F50-based MPC, and datastreams were printed on an engineering level InfoPrint 4000 IR1/IR2 with pre-release V9.0 microcode. In the course of benchmarking datastreams used in prior testing, some measurements were made of Postscript RIP times. These results will be used along with deduced (but not measured) extrapolations to compare and contrast both printing on InfoPrint 4000 464/600 (the 3900 DR1/DR2) and a combination of InfoPrint 4000s and InfoPrint 60s. Limited measurements were also made with traditional IPDS datastreams sent to the InfoPrint 4000 IR1/IR2 from PSF/MVS and PSF/AIX. These allow some comparisons to be made between the InfoPrint 4000 IR1/IR2 and the 3900 DW1/DW2. These comparisons will be of interest to production printing customers. The performance predictions in this FLASH should be used to help properly configure an InfoPrint software and hardware solution and to set customer expectations regarding performance when printing customer applications. WHAT ARE THE ENHANCEMENTS? Before discussing performance, a review of the differences between the "old" and "current" systems is in order. The "old" InfoPrint MPC was either a micro-channel based 39H or a PowerPC PCI-based F30. These two RS/6000s had very similar performance characteristics. To help in the comparison, here is a description of the primary hardware configuration of the F30: 133 MHz Power PC 604 microprocessor with 160 MB main memory, 16 KB L1 cache, 3 4.5 GB SCSI hard drives, a 4X SCSI CDROM, 4/16 Mbps Token Ring adapter, 10 Mbps Ethernet adapter, and an 8 MM tape drive. The new InfoPrint MPC is a 4-way PowerPC PCI-based F50. Each processor in the complex is an 166 MHz PowerPC 604e with speed-matched X5 cache (32 KB L1 and 256 KB L2 cache per processor). The system has a total of 512 MB main memory, 4 4.5 GB SCSI hard drives, 100 Mbps FDDI adapter, 10/100 Mbps Ethernet adapter, 4/16 Mbps Token Ring adapter, 8X SCSI CDROM, 28.8 external modem (for remote diagnostics), and a 4 MM tape drive. The F50 also has three individual PCI buses for improved data throughput between the microprocessors and the I/O subsystem. Further information on both the F30 and F50 is available on the Internet at http://www.rs6000.ibm.com/hardware/workgroups/f30.html and http://www.rs6000.ibm.com/hardware/enterprise/f50.html . On the printer side, the primary differences between the InfoPrint 4000 464/600 (3900 DR1/DR2) and the InfoPrint 4000 IR1/IR2 are a more advanced control unit with additional memory (512 MB versus 256 MB), switchable resolution (the IR1/IR2 can print at either 480 dpi or 600 dpi), support for both POD and traditional IPDS print jobs, and the addition of 100 Mbps FDDI attachment when run in POD mode (the 464/600 used 16 Mbps Token Ring). PERFORMANCE IMPROVEMENTS IN POD MODE Like the InfoPrint 4000 464/600, the InfoPrint 4000 IR1/IR2 can print up to 464 2-up, 8.5"x11" impressions per minute. Actual throughput can be affected by the mix of jobs and processes active on the MPC, the type and complexity of the input datastream (Postscript or TIFF), the level of compression achieved in the resulting IOCA Group 4 datastream (higher levels of compression perform better in general), the presentation and orientation of data pages, the duration of individual print jobs (number of pages per copy and the number of copies printed in a single run), and (when in production mode) the resolution of the printer (some jobs printed at 600 dpi may not perform as well as the same jobs printed at 240 dpi on a 3900 DW1/DW2 or InfoPrint 4000 ID1/ID2). While there will always be exceptions, the following generalizations summarize the improvements in InfoPrint performance: 1. RIP IMPROVEMENTS o Each microprocessor in the F50 4-way complex is faster than the microprocessors in the 39H and F30. Since Postscript RIPs are CPU-intensive, an immediate benefit of the faster chip is that Postscript RIP performance has increased by a factor of 1.3 to 2. This means file RIP time on the F50 will be 70% to 50% of the time seen on the 39H or F30. o As mentioned in the prior flash, attempting to run both InfoPrint 4000 and InfoPrint 60 Postscript RIPs on a single MPC caused RIP times to essentially double, since both RIPs had to share a single microprocessor. This delay is effectively eliminated with the F50 MPC, since the two RIPs can be run simultaneously on separate microprocessors. Combined with the performance improvements mentioned in the prior item, simultaneous RIPs will complete in as little as 25% of the time seen on the 39H or F30. o TIFF transforms will also benefit in two areas: the increased microprocessor performance will give quicker RIP times, and having four microprocessors means multiple TIFF transforms can potentially run simultaneously. 2. OVERLAPPING RIP AND PRINT IMPROVEMENTS The prior flash cautioned customers to carefully schedule when files were RIPped and printed, especially to InfoPrint 4000 464/600 or to a combination of the InfoPrint 4000 464/600 and up to three InfoPrint 60s. The reason for this recommendation was to avoid delays in downloading data to the InfoPrint 4000, since both RIP and print processes were competing for time on the single microprocessor. Since the F50 has 4 microprocessors, and the RIP process is primarily CPU-bound while the downloading process is primarily I/O-intensive, up to two RIPs can be active simultaneously with downloading and printing to an InfoPrint 4000 IR1/IR2 and one or more InfoPrint 60s. This means customers have more flexibility in scheduling both RIPping and printing. As a general recommendation, it is still appropriate to submit jobs to InfoPrint 4000s using the RIP-AND-HOLD option. For customers with a single InfoPrint 60, RIP-AND-PRINT is appropriate. Customers with multiple InfoPrint 60s will get the highest throughput by using a combination of RIP-AND-HOLD and submitting jobs from Operations during non-printing periods. Pre-RIPping is the ONLY way to enable SIMULTANEOUS printing of Postscript jobs on multiple InfoPrint 60s, since only a single InfoPrint 60 Postscript RIP can be active in the MPC at a time. 3. PRINTING IMPROVEMENTS o The most significant improvement in the printing process is the reduction in the time to transfer data between the MPC and the new InfoPrint 4000 IR1/IR2. This improvement is a direct result of the higher processing power in both the MPC and the printer, and the addition of FDDI attachment. The prior InfoPrint solution used a 16 Mbps Token Ring for data transfer. This provided sustained transfer rates between .5 and 1 MBps in most cases. The new system has vastly increased data rates to between 4 and 5 MBps in most cases. As a result, spooling and saving page rates are significantly faster than before. Laboratory testcases exhibited a range of 3 to 15 times improvement in saved-page rates for pre-RIPped jobs. As an example, one job that previously took 112 seconds to "save" pages to the 3900 DR1/DR2 took only 23 seconds to save pages on the InfoPrint 4000 IR1/IR2. o "Proof Mode": In the previous solution, using proof mode was not encouraged, since this often resulted in backhitching. The recommendation was to do all production work in "saved-page" mode, since this eliminated clutching in most cases (which means most jobs in saved-page mode printed at rated speed once printing commenced). The hardware and microcode improvements in the June solution have significantly improved the performance of "proof" (also called non-saved-page) mode. In most cases, jobs printed in proof mode will perform about as well as they previously did when printed in saved-page mode. This means many jobs can now print at rated speed when sent in proof mode. However, complex jobs will still clutch in proof mode. IBM recommends that commercial print on demand applications continue to be printed using saved-page mode. o "Saved-Page" Mode: Performance in saved-page mode has also greatly improved. Printing rates after pages have been saved are up to 150% faster for jobs that previously printed below rated speed after pages had been saved and buffered. Combined with the large reduction in spooling and saving pages (reductions of 3 to 15 times have been seen in limited benchmarks), the number of copies of a print job required to eliminate the inter-job saving page delay has been significantly reduced. Previously, more than 5 copies were generally required to eliminate inter-job delay. This has been reduced to 2 copies in most cases. With these improvements, there is little benefit to running jobs in proof mode. Saved-page mode can be used exclusively in most cases, even with a mix of production (many copy) and proof (one or two copy) jobs. PERFORMANCE IMPROVEMENTS IN IPDS MODE Limited performance tests were made with the InfoPrint 4000 IR1/IR2 in IPDS mode. Tests were done with both PSF/MVS (ESCON attached) and PSF/AIX (FDDI attached). These tests show the InfoPrint 4000 IR1/IR2 generally performs as well as or better than the 3900 DW1/DW2 in both 480 and 600 dpi modes (printing 240, 300, and 600 dpi jobstreams). NOTE: Both the 3900 DW1/DW2 and the InfoPrint 4000 ID1/ID2 can print up to 708 2-up 8.5"x11" impressions per minute. Performance comparisons between these printers and the 464 ipm InfoPrint 4000 IR1/IR2 are made as follows: whenever the 3900 DW1/DW2 and InfoPrint 4000 ID1/ID2 exceed 464 ipm, the InfoPrint 4000 IR1/IR2 is said to perform "as well as" the comparison printers when it prints at 464 ipm. For complex benchmarks that fall below 464 ipm on the 3900 DW1/DW2 and InfoPrint 4000 ID1/ID2, the InfoPrint 4000 IR1/IR2 is said to perform as well as the comparison printers when its print rate matches the rate on the comparison printers. The major improvements production printing customers will see are improved resolution (480 or 600 dpi versus 240 or 300 dpi), and (for complex jobs) a reduction or elimination of any backhitching seen on 3900 DW1/DW2 or InfoPrint 4000 IR1/IR2. NOTE: Some applications which print at rated speed on IBM 3900s and InfoPrint 4000s at 240 dpi resolution may print below rated speed on the InfoPrint 4000 IR1/IR2. SUMMARY The InfoPrint 4000 IR1/IR2 and the new F50-based MPC provide significant performance improvements over the original InfoPrint Solution composed of either a 39H or F30 MPC and an InfoPrint 4000 464/600 (3900 DR1/DR2) Duplex Printing System. For customers upgrading existing systems to the June 1997 MPC and the InfoPrint 4000 IR1/IR2, the major visible performance improvement will be the significant reduction in the delay between the printing of the header page for a job and the printing of the first data page. The elimination of delays between jobs when at least 2 copies of an average job (meaning each job contains 100 or more pages) are printed will also be noticed. Improvements also apply to attached InfoPrint 60s, enabling multiple InfoPrint 60s (up to 3) to be attached in conjunction with an InfoPrint 4000. While performance measurements have not yet been made for attaching two InfoPrint 4000 IR1/IR2s to a single MPC, the performance improvements discussed above suggest that two printers per MPC is a much more viable option than with the previous solution. Except as noted in the discussion above, the recommendations and observations made in FLASH 9651 continue to apply. Careful review should still be done before recommending attaching more than one InfoPrint 4000 IR1/IR2 to an MPC.