cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ivan Kudryavtsev <kudryavtsev...@bw-sw.com>
Subject KVM Host overprovisioning
Date Sun, 10 Sep 2017 06:37:15 GMT
Hello, community.

During the last years Linux kernel got some interesting features like KSM,
ZSWAP, ZRAM. Hardware also steps forward and we see Intel 3d xpoint,
extremely fast SSD drives with m.2 and PCI-E interfaces.

These facilities enable potentially interesting use of overcommited RAM for
hosts. According to IBM's investigations Zswap with LZ4/ZBUD increases
virtual RAM on 40%.

I investigated current Apache CloudStack memory overcommitment capabilities
and they mostly affect VM's RAM by utilizing ballooning and I think it's
not what necessary to open new facilities. There are many cases which can
utilize ZSWAP and fast swap devices to efficiently provision more RAM than
presents.

I suppose, CloudStack Agent for KVM can have configured parameter which
"mangles" RAM reported. From the other hand it can be done by implemented
host properties on server side. I tried manual increasing in host table:

update host set ram=ram * 1.4 where id=1;

and it seems until the next host stats update it works as expected. I think
this workaround is useful, but it's better to have the function in core
like standard.

Let me know what you think about it, it might be I don't understand
something and ACS already has it in place? I also would like to hear your
thoughts on ZSWAP usage in practice.

-- 
With best regards, Ivan Kudryavtsev
Bitworks Software, Ltd.
Cell: +7-923-414-1515
WWW: http://bitworks.software/ <http://bw-sw.com/>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message