Thanks to the following:
Dave Newbold <phdmn_at_siva.bris.ac.uk>
Alan Rollow <alan_at_nabeth.cxo.dec.com>
Pulak Rakshit <pulakr_at_cableol.net>
John Kinsella <johnk_at_ucdmath.ucdavis.edu>
Paul E. Rockwell <rockwell_at_rch.dec.com>
Eugene Chu <chu_at_musp0.Jpl.Nasa.Gov>
Jim Belonis <belonis_at_dirac.phys.washington.edu>
Hellebo Knut <bgk1142_at_bggfu2.nho.hydro.com>
Murat Balci <balci_at_baum01.ege.edu.tr>
>Considering that the RAM = swap space size = 64 Mb and the CPU load (in
>the form of processes) at that particular time is constant, would ...
>1.) Doubling RAM size (from 64 to 128 Mb) and maintaining the swap space
size (64 Mb)
> have the same, lesser or greater performance than..
>2.) Maintaining RAM size (64 Mb) and doubling the swap space size (64 to
>128 Mb)?
Most of the respondents suggested that increasing of RAM size HAS BETTER
results (greater performance) than increasing swap space size. One
consideration is the fact the RAM I/O is faster than the disk.
Some respondents also suggested that the swap space size should be
ranging from 0.5 to 3 times greater than the size of RAM size.
Alan Rollow discusses more...
----------------------------
Depending on how you increase the amount of swap space it
will have little or no affect on the performance of the
system as a whole. It may affect the performance of indiv-
idual processes in a binary fashion, allowing them to run
where they wouldn't before due to a lack of virtual memory.
The increase of physical memory, may have a significant
affect on the performance, since it is sure to reduce the
amount of paging or swapping. If eager page/swap space
allocation is used, an increase in the amount of physical
memory will have to be accompanied by an increase in the
amount of page/swap space.
Details...
Since V1.3 Digital UNIX (then DEC OSF/1) would interleave
page/swap allocations among all the available page/swap
space disks. This would effectively stripe the data
among the available devices, offering better performance.
How much better would depend on the configuration of the
disks. For example, having two disks on two seperate
busses should offer better I/O performance than a two
disks on the same bus.
The gain of having a better I/O configuration is usually
vastly overshadowed by increasing the amount of memory,
which will reduce paging.
Digital UNIX offers two page/swap space allocation methods;
eager and lazy. The default is eager and reserves page/swap
space when virtual memory is allocated. The space isn't
actually allocated until needed to write out pages. The
reservation ensures that there will enough page/swap space.
This method limits the amount of virtual memory on the system
to the amount of page/swap space available.
The alternative, lazy, doesn't reserve the space. It too
is allocated when needed. If insufficient space exists
when the allocation is made, processes are killed to free
up space. The assumption is made that the current process
should be allowed to continue running no matter what. In
extreme cases important processes like the X server will
eventually be killed off to get enough page/swap space.
The advantage of lazy page/swap space allocation is the total
virtual memory supported by the system is the sum of physical
memory and the page/swap space. The disadvantage is that a
short fall in the amount of page/swap space can causes processes
to be killed.
Some people have claimed to see better performance when using
lazy page/swap allocation, but this seems unlikely since the
only thing that is supposed to be different is the reservation
of page/swap space. While the reservation of space does take
a little extra time, it should be insignificant. Of course,
once processes start dying off, there will be a performance
change as more physical memory becomes available.
Paul E. Rockwell has this to say:
---------------------------------
> 1.) Doubling RAM size (from 64 to 128 Mb) and maintaining the swap space
> size (64 Mb)
If you're running in the default swapping mode ("eager" allocation), then
this is a bad idea. You can never allocate more than 64MB of anonymous
virtual memory, so you're limiting the size of programs running on your
system.
In either swap mode, it's a bad idea because your swap partition can't hold
a crash dump of the entire memory of the system.
> 2.) Maintaining RAM size (64 Mb) and doubling the swap space size (64 to
> 128 Mb)?
No change in performance. The only thing this allows you to do is to
potentially make performance worse. by oversubscribing memory.
The only thing that will help a system that's starving for physical memory
resulting in lots of paging is memory. Additional swap spaces on other
disks may help the disk performance involved in swapping, but it's not
as big a gain as getting more memory.
Nothing helps virtual like real.
>Also, could the Alpha 2000 Model 300 SIMMS (RAM) be replaced with (8)
>eight 72-pin (not sure)16 Mb PC SIMMS (totaling 128 Mb)?
Most of the responses indicated that the Alpha RAMs cannot be replaced
with ordinary PC SIMMS.
Bonn
:)
Received on Tue Aug 29 1995 - 20:22:33 NZST