<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div class="moz-cite-prefix">On 03/16/2014 06:33 PM, Brian Miller
wrote:<br>
</div>
<blockquote cite="mid:532626BE.9030301@gmail.com" type="cite">
<pre wrap="">This isn't exactly based on research, but my current practice is to
provide a small swap space on my virtual machines (say, 500 MB), and
then give the guest enough memory to run the application. </pre>
</blockquote>
<br>
Exactly ... give the guest enough memory to run the application. <br>
<br>
I run most of my virtual Linuxen with <u>no swap space</u> at all.<br>
<br>
<blockquote cite="mid:532626BE.9030301@gmail.com" type="cite">
<pre wrap=""> If the server
writes a small amount to swap, I don't worry about it. But if swap
keeps growing, I generally increase the amount of RAM. Unless it looks
like it's caused by a memory leak. Then I tell the application admins
to set a schedule to re-start their application. I'll even offer to set
up a notification email to tell them when swap gets to 75% full, if they
want.
</pre>
</blockquote>
<br>
Also keep in mind that <u>occupied</u> swap space is not as "heavy"
as <u>active</u> swap space. If a virtual server dips into swap
space when getting started, no sweat. The thing to avoid is moving
pages between memory and disk during normal operation. <br>
<br>
There's no getting around some tuning of guest memory size. Define
guest memory as small as possible to avoid paging/swapping.<br>
<br>
It's good to evaluate swapping/paging capabilities of the host
(hypervisor). The z/VM system is really good about paging. As I
recall, some versions of VMware don't actually perform swapping of
host memory. (I really haven't looked at the others on this point at
all.) Ironically, it's better for the host to do it than for the
guest to do it. <br>
<br>
-- R; <><<br>
<br>
<br>
<br>
</body>
</html>