edhew@xenitec.on.ca (Ed Hew) (01/26/91)
In article <1991Jan24.194219.6604@world.std.com> steveo@world.std.com (Steven W Orr) writes: >I am about to decide to buy a 1.2G Seagate Wren 7 SCSI drive. We are very pleased with our 660meg Wren 6 SCSI boot drive. > The >question has been raised that I might not be able to take advantage >of the space on the disk because each physical device has a limited >number of inodes. While there are other (solved) problems that might prevent you from using all the cylinders of very large drives, lack of inodes is not a major factor. Lack of inodes within a given filesystem can be a headache (like in /dev/news), but that's a different problem. > Could someone explain the relationship between the >inodes per disk (which I think is called a file system) and the >inodes per partition? If there is a max can I configure it or is it >fixed? All the following "as of the last time I checked". First note that SCO UNIX 3.2.* can support 2 SCSI host adapters each of which can support 2 SCSI hard drives. I'm told that this limitation only exists due to the SysV maximum of 4 availabie minor device numbers for individual hard drives. Each hard drive will support up to 4 partitions. Each partition will support up to 7 filesystems plus an entry for the device (eg, active_UNIX_partition=/dev/hd0a). The active UNIX partition will house the mandatory swap area and the suggested 'recover' area. You may also wish to have a scratch filesystem. This gives you 4+7+7+7 filesystems available per drive (the first filesystem on the active partition will be root). Each filesystem will support up to 64K (that's 16*16*16*16=65536) less 2 (reserved) inodes per filesystem. One free inode gives you the ability to "add" one file, assuming you have some free diskblocks for the "optional" data. :-) By default about 25% of the free space in each filesystem is allocated to the inode table. That adds up to a lot of potential inodes if the filesystems are each able to be physically large enough to get the max # of inodes. A rough guestimate suggests that you'd hit the optimal SysV inode max with a 5-ish-Gig drive. Now it's up to you to determine the optimum layout. You can use the -v option to divvy to access "virtual" (non-active) partitions. I would recommend: man divvy man fdisk man mkfs > Also, does anyone have anything good or bad to say about this >particular drive? The best price I've found so far is $2800 for the >bare drive. I'm also interested in speculation about the prices >dropping substantially in the near future. History suggests a constant decline in the $/meg storage cost. I don't see technology holding back a continuation of this trend. There are of course bumps in the graph. Buy when you _need_ it. >Email me, and if there's enough clamor I'll summarize. I'll post the answer to this as there are likely others wondering what the bottom line is. I'll email you the illustration using our site as an example. Again, corrections and elaborations from those more knowledgeable are always welcome. >Steven W. Orr steveo@world.std.com uunet!world!steveo -- Ed. A. Hew <edhew@xenitec.on.ca>, XeniTec Consulting Services or if you're really stuck: ..!{watmath|lsuc}!xenitec!eah [sco.opendesktop newsgroup <=> mlist gateway maintainer]
ires@kaspar.UUCP (Bruce R. Larson) (01/28/91)
>In article <1991Jan24.194219.6604@world.std.com> steveo@world.std.com (Steven W Orr) writes: >>I am about to decide to buy a 1.2G Seagate Wren 7 SCSI drive. >> >>The question has been raised that I might not be able to take advantage >>of the space on the disk because each physical device has a limited >>number of inodes. > Here are my two cents worth. In SysV/386, up to Rel 3.2.2 at least, the following are true: *) there are a maximum of 65,536 inodes per filesystem or division *) there are generally five (5) device names, as Ed Hew mentioned, available for partitions to be associated with filesystems. [NOTE: Ed also mentioned that if you have a SCSI hard disk, then it may be possible to reclaim SCO's `recover' area for use as an additional filesystem, however I'd try it myself before recommending it to a client. In any event, if you succeed you will have six filesystems instead of five.] *) the default ratio of 512-byte blocks per inode for the mkfs command is eight (8), and news filesystems are usually made with four blocks per inode (because news consists of lots of small files). This summarizes to a max of 1.28GB on a single disk with five 256MB filesystems created using mkfs' default blocks:inodes ratio. Since a 1.2GB disk should format down to to about 980MB or so, it is certainly possible to cover the disk with five filesystems, even if one is a 128MB partition for news, but that doesn't mean it is wise to do so. I prefer to configure a disk with a filesystem structure that promotes good performance and easy maintenance, so a 1.2GB disk would be impractical for me. I'm too accustomed to the luxury of separated root, tmp. usr, u and usr2 filesystems to compromise their existence for the sake of getting a monster capacity hard disk. I'd rather have two merely huge disks, like the Wren VI FH or the soon-to-be-released 12ms 24mz Wren Runner II (a real smoker, this one), and enjoy the benifits of increased capacity and performance while maintaining the safety of separated fileystems. However, if someone gave me a 1.2GB disk, here is how I would probably slice it up (add salt to taste): * small root filesystem (/) 35MB to 75MB * swap space 15MB to 80MB * usr filesystem 150MB to 256MB * src filesystem 256MB * news filesystem 128MB (4 blocks per inode) * u filesystem the rest I decided to post this instead of mailing to Steve (or calling him) because this seemed to be of general enough interest. Bruce Bruce R. Larson | Integral Resources, Milton MA | /* Temporary address change below */ | Uucp: ..!world.std.com!kaspar.uucp!blarson | Internet: blarson@cs.umb.edu | -------------------------------------------