Re: NFS performance

[Lukas Beeler <lukas.beeler@xxxxxxxxxxxxxxxx>]

On Fri, Aug 14, 2009 at 20:10, Pat Barber<mboceanside@xxxxxxx> wrote:
> Doing a data transfer "does not" cause a CPU to be
> bound.

CPU bound = limited by CPU. Which is what you see on the very small
machines when doing large NFS or FTP transfers (100% CPU load caused
by the FTP process).

> A very narrow data path with limited memory causes
> the process to be slow but the processor is doing
> VERY little work.

Gigabit Ethernet is 125 Megabyte/s. Significantly more than the IBM i
entry level tape drive, the DAT72. It is rated at 3 Megabyte/s.

SLR60 2.5" drives are rated at 4 Megabyte/s, which is what most of the
Power 520 i've installed back then shipped with.

Most model 170ies had a 100mbit/s NIC, which can provide up to 12.5
Megabyte/s, which is significantly more than the MLR3 had (2
Megabyte/s).

> The data path is the restriction. Doing massive copying
> on a network of twisted pair is both slow and silly.

It's slower than LTO4, i'll give you that (LTO4 is 120 Megabyte/s
uncompressed, but through compression up to 240 Megabyte/s may be
achieved).

> Most any tape drive made in the last 10 years is
> much faster than a "network transfer".

No, only high-end, high-performance tapedrives like the LTO line of
products are fast. Most other tape options are rather slow, though
some of them can beat 100mbit Ethernet.

>  Your comments on the speed of older machines shows a
> real lack of experience in these computers.

Well, i'll give you that. I've mostly migrated away from these
machines or had to debug issues.