Dell 3-Node AMD DCS6005
- Thread starter Xeppo
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Opteron 2419 EE - here's a link to the benchmark I ran a few days ago...I believe you need low profile cards from what ive read.
Anyone whos bought one form that same seller, that i did, can you confirm which CPU's are in these?
I just ordered one of these for my home vSphere lab.
Has anyone ordered a second power supply for their unit? Mine only comes with one. I'd read online that many of these units were designed to only work with a single power supply. Just curious if anyone had tried a second one.
Has anyone ordered a second power supply for their unit? Mine only comes with one. I'd read online that many of these units were designed to only work with a single power supply. Just curious if anyone had tried a second one.
i talked to the seller today - they said a 4th board and 2nd ps was not possible... But im betting some of the things that are standard in the non dcs cases are missing and could probably be purchased separately..
something like this might be missing... New Dell PowerEdge C6100 1st 2nd Power Distribution Boards TJ77Y YJ9Y6 | eBay
dont have mine yet so couldnt say..
something like this might be missing... New Dell PowerEdge C6100 1st 2nd Power Distribution Boards TJ77Y YJ9Y6 | eBay
dont have mine yet so couldnt say..
I haven't ordered a 2nd power supply. I am very happy that I can get this much value with a power supply that I can power from a regular wall outlet.
I just ordered one of these for my home vSphere lab.
Has anyone ordered a second power supply for their unit? Mine only comes with one. I'd read online that many of these units were designed to only work with a single power supply. Just curious if anyone had tried a second one.
If possible, as a suggestion, try installing esxi 5.1 or 4.1 one one of the other drives and set that drive as the boot drive and see if the powering off issues continue. You could also try installing another OS instead of esxi altogether on one of the other drives and setting the bios to boot from it in the node having the power issues. That way, you can rule out the OS as the possible cause.
Im running Esxi 5.5
So do you suggest 1 virtual processor per physical core ratio?
I've had bad luck with Dell SIPs and PS/2 to USB converters on Dell Poweredge servers, but these aren't really Dell servers, they are Tyan, so it's worth a shot... Just save your receipt.
As for your dozen servers w/ 2 Gigs of RAM on one 1TB drive, I'd seriously consider using multiple spindles in a RAID0 (stripe) array to get better performance. Also, it depends on what the servers are doing - the VM would have normal OS disk activity, application disk activity, and if memory is too low, paging activity also. I'd ensure that each VM has enough memory to avoid paging activity as much as possible.
Hypervisor hosts need very little RAM compared to regular servers, consider bumping each VM to 3 or 4 Gigs, you only need a couple gigs for ESXi or whatever.
I'm curious being new to virtualization. If a particular software install recommends 6GB of RAM and you wanted to install this on a guest vm, would you give the vm the 6GB of RAM or go with 4GB given the lowered RAM needs of a vm?
"Hypervisor hosts need very little RAM compared to regular servers, consider bumping each VM to 3 or 4 Gigs, you only need a couple gigs for ESXi or whatever."
"Hypervisor hosts need very little RAM compared to regular servers, consider bumping each VM to 3 or 4 Gigs, you only need a couple gigs for ESXi or whatever."
I was commenting about VMs in the abstract, if a particular application calls for, say, 6 Gigs of RAM if installed on actual hardware, you should allocate that same amount of a RAM to the VM it runs inside.I'm curious being new to virtualization. If a particular software install recommends 6GB of RAM and you wanted to install this on a guest vm, would you give the vm the 6GB of RAM or go with 4GB given the lowered RAM needs of a vm?
"Hypervisor hosts need very little RAM compared to regular servers, consider bumping each VM to 3 or 4 Gigs, you only need a couple gigs for ESXi or whatever."
Some OSs function fine with 2 Gigs of a RAM, other will page/swap like crazy with less than 4 Gigs of RAM.
Simply running software inside a VM doesn't alter the requirements of the software with regard to system resources.
I used to think a one-to-one Virtual to Physical CPU ratio was best, but that had me running (for example) seven single-core Windows Server instances on a dual quad core host (leaving one core for the hypervisor), but further investigation tells me that I would likely see better performance if I assigned two CPU cores per VM, over-subscribing the CPUs by 100%, since that has each VM running with two cores, a much happier place for most VMs. This is theoretical, based on reading I've done - I've very little experience with this so far.So do you suggest 1 virtual processor per physical core ratio?
The interesting question for me is hyper threading on Intel CPUs - with hyper-threading enabled the hyper visor sees 2x as many cores, but half of those cores are poor performers, with lower throughput that a 'full' core... Is it better to enable hyper-threading and allocate more CPUs per VM, or is it better to disable hyper-threading and assign correspondingly fewer CPUs per VM? My gut tells me the former (enabling Hyper-threading and assigning more CPUs per VM improves performance), but I'm not so sure...
My best advice is to not be afraid to over-subscribe your CPUs two virtual CPUs per actual CPU is probably a good rule of thumb, but a 3:1 ratio is probably fine in a pinch, based on workload.
I ordered SPARE power supplies, at $60 for two, it is cheap insurance - if the PS took a hit next year, where would I find a replacement PS for this chassis?I just ordered one of these for my home vSphere lab.
Has anyone ordered a second power supply for their unit? Mine only comes with one. I'd read online that many of these units were designed to only work with a single power supply. Just curious if anyone had tried a second one.
Thanks Ken. This is helpful. I take it that when you deploy a guest vm that was pre-configured when it was made available that the resources needed to support the VM are already known during the deployment process.
I was commenting about VMs in the abstract, if a particular application calls for, say, 6 Gigs of RAM if installed on actual hardware, you should allocate that same amount of a RAM to the VM it runs inside.
Some OSs function fine with 2 Gigs of a RAM, other will page/swap like crazy with less than 4 Gigs of RAM.
Simply running software inside a VM doesn't alter the requirements of the software with regard to system resources.
Thanks for such a detailed reply. This is indeed very helpful and gives me a good foundation to start from.I used to think a one-to-one Virtual to Physical CPU ratio was best, but that had me running (for example) seven single-core Windows Server instances on a dual quad core host (leaving one core for the hypervisor), but further investigation tells me that I would likely see better performance if I assigned two CPU cores per VM, over-subscribing the CPUs by 100%, since that has each VM running with two cores, a much happier place for most VMs. This is theoretical, based on reading I've done - I've very little experience with this so far.
The interesting question for me is hyper threading on Intel CPUs - with hyper-threading enabled the hyper visor sees 2x as many cores, but half of those cores are poor performers, with lower throughput that a 'full' core... Is it better to enable hyper-threading and allocate more CPUs per VM, or is it better to disable hyper-threading and assign correspondingly fewer CPUs per VM? My gut tells me the former (enabling Hyper-threading and assigning more CPUs per VM improves performance), but I'm not so sure...
My best advice is to not be afraid to over-subscribe your CPUs two virtual CPUs per actual CPU is probably a good rule of thumb, but a 3:1 ratio is probably fine in a pinch, based on workload.
I assume you mean, for example, importing a pre-built VM, say from Microsoft or Turnkeylunux.com? In those cases the VM configuration should be 'baked in' to the VHD/virtual disk set to what the vendor felt was appropriate.Thanks Ken. This is helpful. I take it that when you deploy a guest vm that was pre-configured when it was made available that the resources needed to support the VM are already known during the deployment process.
Of course, the settings can be adjusted up or down, but they should be ready to go, except for environment specific settings like NIC assignment.
You should always start with least possible amount of VCPU per VM, what i've found is that using SSD makes a huge difference in performance,and that in turn reduces the amount of RAM needed to get things done.
Over subscribing VM hosts - you have to consider NUMA locality, etc. It doesn't make sense to have a VM that is idling core2 90% of the time, since the scheduler will have to wait for 2 slots to schedule that VM to run, where as one vcpu will always get a slot since it is the lowest possible size.
I tend to keep "like' vm's on the same physical host. All 1 VCPU together, all 2 VCPU together, and all 4 VCPU (by themselves). Also don't forget to leave enough reserve for the hypervisor. It can use up a full core depending on what it is doing at times!.
For a quad-core cpu, I might schedule 3 1-vcpu vm's and leave 1 for the hypervisor.
For a 6-core cpu, I might schedule 1 4vcpu VM and a simple 1 VCPU vm.
It seems counter-intuitive, but hypervisors are aweful when it comes to fair-sharing on simple setups (no SAN/no SR-IOV). They don't share resources very well at all!
It's why my DL380 has two LSI megaraid controllers (prefer to have had 4 9266-4i rather htan 2 9266-8i) since then I could dedicate one raid controller to a VM, no sharing at all.
Bottom line: baseline your performance with 1 VM per host, then add more VM's and watch the performance plummet!
Over subscribing VM hosts - you have to consider NUMA locality, etc. It doesn't make sense to have a VM that is idling core2 90% of the time, since the scheduler will have to wait for 2 slots to schedule that VM to run, where as one vcpu will always get a slot since it is the lowest possible size.
I tend to keep "like' vm's on the same physical host. All 1 VCPU together, all 2 VCPU together, and all 4 VCPU (by themselves). Also don't forget to leave enough reserve for the hypervisor. It can use up a full core depending on what it is doing at times!.
For a quad-core cpu, I might schedule 3 1-vcpu vm's and leave 1 for the hypervisor.
For a 6-core cpu, I might schedule 1 4vcpu VM and a simple 1 VCPU vm.
It seems counter-intuitive, but hypervisors are aweful when it comes to fair-sharing on simple setups (no SAN/no SR-IOV). They don't share resources very well at all!
It's why my DL380 has two LSI megaraid controllers (prefer to have had 4 9266-4i rather htan 2 9266-8i) since then I could dedicate one raid controller to a VM, no sharing at all.
Bottom line: baseline your performance with 1 VM per host, then add more VM's and watch the performance plummet!
fnc1, this is what I ordered from Amazon.
Amazon.com : SanDisk Cruzer Fit SDCZ33-008G-B35 8 GB USB 2.0 Flash Drive - Black : Electronics
Amazon.com : SanDisk Cruzer Fit SDCZ33-008G-B35 8 GB USB 2.0 Flash Drive - Black : Electronics