Article 18970 of comp.sys.sun.hardware: Xref: mri-gw comp.periphs.scsi:23436 comp.sys.sun.hardware:18970 Path: mri-gw!psinntp!psinntp!psinntp!unvax.union.edu!news.sprintlink.net!howland.reston.ans.net!wupost!intron!ht1 From: ht1@intron.wustl.edu (Haroon Taqi) Newsgroups: comp.periphs.scsi,comp.sys.sun.hardware Subject: Re: Mixing slow SCSI and FAST SCSI-2 Followup-To: comp.periphs.scsi,comp.sys.sun.hardware Date: 3 Oct 1994 17:14:08 GMT Organization: Washington University in Saint Louis, MO USA Lines: 67 Message-ID: <36pe50$4c6@wupost.wustl.edu> References: <36fbmd$4u7@wupost.wustl.edu> NNTP-Posting-Host: intron.wustl.edu X-Newsreader: TIN [version 1.2 PL2] First of all thanks to following for responding: dplatt@3do.com jgp@ceas.rochester.edu erich_oetting@stortek.com jdd@cdf.toronto.edu tonico@kaiwan009.kaiwan.com I have summarized their responses below which should help other SCSI novices like myself: >1) Do all SCSI-2 disks support FAST SCSI transfer rates of 10MB/s? (I am >not referring to the wide-scsi disks).... No. All SCSI-2 mandated was a cleaned up version of regular SCSI. Most of the quality vendors were able to pencil in a 2 on their existing drives when the new standard came out since they were already using the generally accepted solutions to SCSI-1's problems. Fast (and Wide) is an optional part of SCSI-2. Morever, a device might be "SCSI-2" and not be willing to negotiate a net transfer speed of >8 MB/sec, for example. Figure that "SCSI-2" means "between 5 and 10". >2) What is the maximum throughput on the SCSI bus if we mix slow SCSI (max >througput 5MB/s) disks on the same chain with fast SCSI-2 disks (10MB/s >max) on a machine that has a Fast SCSI-2 disk controller (e.g. a Sparc >... >However the local Sun Engineer claimed that you should get the maximum >throughput for that device for each device. In other words, you should get >10MB/s data transfer from the fast disk and 5MB/s from the slow disk. The host negotiates max transfer rate with each device separately, so the local Sun engineer is correct, if everything is implemented properly. Initiator and target negotiate the sync transfer speed during the IDENTIFY phase at the beginning of each transfer. In the presence of non-optimal cabling, the negotiated rates might be less than the max permitted (e.g. only 8MB/s if your bus is too long). >3) Finally, is it recommended to put fast SCSI and slow SCSI disks on the >same chain? If not, what is the alternative, esp. when we have older disks >and newer equipment that needs to go on the same machine? Works fine as long as you keep the following issues in mind: - With single-ended SCSI (ie the normal non-differential cabling) running at Fast speed is marginal. The shorter the cable and the fewer devices on it the faster the possible data rate is. If you are planning on hooking 4 shoeboxes together with 3ft cables you probably aren't going to get top speed from the devices. - The slow devices will have an impact on overall throughput since they will tie up the bus for twice as long as a Fast device would. You could get better performance by adding another SCSI controller but in a lot of cases the extra performance is not worth the cost. - Lots of old SCSI stuff might use inferior cable or connectors since 10Mbyte/s didn't exist when it was created. - Fast SCSI is much pickier about termination and cabling, especially on suns. If you run into problems, your could get another separate controller. - When in doubt: use high-quality ribbon cable and gold-plated connectors, keep the connectors 30 cm apart on the cable, don't put any "stubs" on the bus, and use an active terminator at each end. - Scsiinfo (ftp.cdf.toronto.edu:/pub/scsiinfo/*) or prtconf -v (on Solaris 2.x) will tell you what transfer rates are in effect for each device. Haroon WashU Genetics Dept. Email: ht1@genome2.wustl.edu