AIX

I have questions about lslv attributes.

- The higher "IN BAND" %, the better intra-allocation policy met, correct?
- Edge is the fastest I/O and Inner-Edge is the slowest (edge : middle : center : inner-middle : inner-edge), correct?
- Is it possible to change DISTRIBUTION manually to get better performance?

Code:

dprod_0000169:N/A
PV                COPIES        IN BAND       DISTRIBUTION
hdisk135          512:000:000   22%           115:115:115:115:052
dprod_0000006:N/A
PV                COPIES        IN BAND       DISTRIBUTION
hdisk19           001:000:000   100%          000:001:000:000:000
hdisk24           004:000:000   0%            000:000:000:004:000
hdisk30           008:000:000   100%          000:008:000:000:000
hdisk31           005:000:000   100%          000:005:000:000:000
hdisk37           195:000:000   58%           000:115:080:000:000
hdisk44           067:000:000   23%           051:016:000:000:000
dprod_0000144:N/A
PV                COPIES        IN BAND       DISTRIBUTION
hdisk71           111:000:000   28%           000:032:000:044:035
hdisk74           084:000:000   0%            048:000:017:000:019
hdisk90           120:000:000   24%           000:029:000:000:091
hdisk94           037:000:000   0%            025:000:000:012:000
hdisk116          128:000:000   39%           000:051:077:000:000
prod_0000096:N/A
PV                COPIES        IN BAND       DISTRIBUTION
hdisk19           001:000:000   0%            000:000:000:001:000
hdisk24           010:000:000   0%            000:000:000:010:000
hdisk29           128:000:000   5%            035:007:051:035:000
hdisk53           041:000:000   63%           000:026:015:000:000
dprod_0000097:N/A
PV                COPIES        IN BAND       DISTRIBUTION
hdisk19           004:000:000   0%            000:000:000:004:000
hdisk24           004:000:000   0%            000:000:000:000:004
hdisk30           008:000:000   0%            000:000:000:008:000
hdisk2            084:000:000   17%           000:015:037:032:000
hdisk50           020:000:000   0%            010:000:000:010:000
hdisk51           022:000:000   0%            022:000:000:000:000
hdisk55           048:000:000   33%           000:016:000:012:020
hdisk81           048:000:000   0%            000:000:048:000:000

First off: it helps if you describe your system so that others understand what you are doing. That you are using AIX and ask questions about the volume manager is not obvious to everybody.

Second: we have a special forum for AIX, where most of our AIX experts read and answer. You would have a better chance to get an answer if you would post such questions there. I will transfer the thread there.

To your questions:

That depends: when a physical volume is aded to a volume group it is sliced into "physical partitions" - blocks of equal size. When space is allocated to a logical volume it is given a collection of such physical partitions as logical partitions: 1 PP per LP if the logical volume is not mirrored, 2 if it is, 3 if it is doubly mirrored. "In band" means that the allocated PPs for LPs in sequence are in sequence too. Consider a non-mirrored LV:

In band:

Code:

PP1 -> LP1
PP2 -> LP2
PP3 -> LP3
PP4 -> LP4
PP5 -> LP5

not in band

Code:

PP3 -> LP1
PP1 -> LP2
PP2 -> LP3
PP5 -> LP4
PP4 -> LP5

If LVs are created and subsquently enlarged it is common to that they become out of band over time. Do a "reorgvg" to get them in band again.

Why "in band" is better is because the read/write-head of the disk will not have to move around so much if reading from logically consecutive disk blocks. With modern SAN storage this becomes obsolete because what appears as "consecutive disk blocks" is virtual anyway.

No. In a typical disk usage pattern data are read and written in a random pattern because several processes do I/O concurrently. In such a random pattern "middle" is the fastest, because the read/write-head can be fastest positioned there from a random position it has had previously.

Yes, you can change the distribution (via the "lmigratepv" command), but it this improves performance depends on the disks used, the usage pattern of your system and its applications and probably some other factors. "Performance" means "fitness for a specific purpose" and without exactly defining what this purpose is and under which parameters it is measured the term becomes meaningless. Is a car faster than a bycicle? On a highway it probably is, but in the center of a city at rush hour the bycicle might well be faster - the answer depends on these parameters.

I hope this helps.

bakunin

Also very important - are these physical disks, or whole spindles from a LUN, or are they stripes across a spindle.

Generally, when coming from a SAN environment, or even iscsi the most important factor is SAN performance configuration and "competing" i/o requests.

In short, if it is not a physical disk I would worry less about the statitistic of in band, or not. More important is whether it is sequential or not (PP1, PP2, ... PP121,PP122) and the i/o is largely sequential.

For random i/o, performance can be heavily affected by the amount of file caching being done by AIX Virtual Memory and/or LVM as well as caching done by the SAN.

Just remember that these statistics were developed for traditional systems with physical (local) storage. They may be next to meaningless in a virtualized environment.