Citrix’s XenServer Engineering’s Performance Team have written a series of articles about how they achieved a density of 500 VMs per host on XenServer 6.2. This is a must read for all those interested in understanding hypervisor performance and architecture and includes information on how our engineering team investigate performance and throughput and internal secrets. Warning contains deep technical insight!
Jonathan Davies’ first blog on the subject outlines event channels in XenServer and improvements our engineering team made. His colleague Felipe Franciosi continues the story covering CPU utilisation of the QEMU process and USB emulation; Jonathan then retakes up the tale in a new blog detailing our position today covering XenServer architectural features such as dom0, blktap2, xenstored, aio requests, consoled and dom0 low memory.
Many of our partners such as Dell and HP are now manufacturing very large servers and being able to support this many VMs ensures customers should not experience any hypervisor limitations. It’s also great for other Citrix products that deploy on the XenServer platform such as NetScaler SDX, XenDesktop, CloudPlatform and VDI-in-a-box.
There’s plenty of marketing information about why VM density matters available but these new blogs are intended for developers and engineers who want to know how we did this. A few weeks ago we put out a developer-to-developer blog on vGPU and we were thrilled by the interest it generated. Our developers regularly post deep technical dives on the xenserver.org blogs and we actively welcome feedback and questions to our developers on the mailing lists or Q&A section of the site. We’re also very interested in writing more of these blogs for developers by developers so please do comment if there are topics you as a developer want to know more about!