You are currently viewing our boards as a guest, which gives you limited access to view most discussions and access our other features. By joining our free community, you will have access to post topics, communicate privately with other members (PM), respond to polls, upload content, and access many other special features. Registration is fast, simple, and absolutely free, so please, <a href="/profile.php?mode=register">join our community today</a>!
Johan, good catch on IO performance for VMMark but you forgot two points:
Quote:
The result is that VMmark with its huge number of VMs per server (up to 102 VMs!) places a lot of stress on the I/O systems. The reason for the Intel Xeon X5570's crushing VMmark results cannot be explained by the processor architecture alone. One possible explanation may be that the VMDq (multiple queues and offloading of the virtual switch to the hardware) implementation of the Intel NICs is better than the Broadcom NICs that are typically found in the AMD based servers
1) Intel NICs are used even for AMD based servers, also it doesn't explain Dunnington slowness. 2) The Nehalem plataform is the only one to support virtualized IO wich still my guess to explain VMMark result, I don't know if VMWare already support SR5690, if so there are a few motherboards with it out there for another good review.
Joined: Sun Sep 23, 2007 1:29 am Posts: 126 Location: Los Angeles, CA
But I think only Intel chipsets have IO/AT, DMA acceleration for their IO subsystems. Turning that on made a big difference in I/O bottlenecks on my old Xeon tests on Linux.
The new AMD chipsets from the Fiorano platform now also finally have support for IOMMU in virtualized environments, so now it could be tested if that really is as a big boost as you say.
Johan, good catch on IO performance for VMMark but you forgot two points:
Quote:
The result is that VMmark with its huge number of VMs per server (up to 102 VMs!) places a lot of stress on the I/O systems. The reason for the Intel Xeon X5570's crushing VMmark results cannot be explained by the processor architecture alone. One possible explanation may be that the VMDq (multiple queues and offloading of the virtual switch to the hardware) implementation of the Intel NICs is better than the Broadcom NICs that are typically found in the AMD based servers
1) Intel NICs are used even for AMD based servers, also it doesn't explain Dunnington slowness. 2) The Nehalem plataform is the only one to support virtualized IO wich still my guess to explain VMMark result, I don't know if VMWare already support SR5690, if so there are a few motherboards with it out there for another good review.
Hi Eduardo,
1) Well Dunnington slowness is more likely a result of the high latency L3 (minor), the lack of EPT (medium) and the extremely low amount of bandwidth for 24 cores:10 GB/s or less than 420 MB/s per core, back to the PII era (major problem!).
1b) Intel NICs are apparantly not used to produce the AMD server VMmark results (I have checked HP's so far)
2) I am not sure. AFAIK, and I am still studying it, the VT-D for Direct I/O only really helps when you use dedidacted I/O devices for VMs. With 102 VMs (and each 6th VM being identical), I can not imagine they actually gave 1/6th of the 102 VMs a dedicated NIC or so.
We are waiting for the Fiorano servers to arrive...
Last edited by Johan on Wed Oct 07, 2009 8:10 am, edited 1 time in total.
Just out of curiosity, how many man-months of effort does it take to perform an analysis like that one?
The hardest article was the first time we came out with vApus mark, which took something like 6 or 7 man months. This one was much shorter (probably a week or 4), but as university people and infrastructure are involved, most of the work went to making sure I can keep my people and the infrastructure (publishing on Anandtech is our "hobby" ;-).
What is the cost of VMware for each server? I understand that VMware uses a 3-D pricing scheme along axes of sockets, cores, and features. Does the cost of VMware ESX overwhelm the performance benefits of using 24 Istanbul cores over 8 Nehalem cores?
Also, I felt as though this chart should have extended from 2GB through 8GB memory modules for each type of system.
1b) Intel NICs are apparantly not used to produce the AMD server VMmark results (I have checked HP's so far)
True, I checked it now, on the other hand Intel submissions use Broadcom based NICs too, in the end, also and looking for other results, changing NICs doesn't appear to change results too much.
Users browsing this forum: No registered users and 1 guest
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot post attachments in this forum
