Virtualization Megathread V2: VMs inside VMs

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > Virtualization Megathread V2: VMs inside VMs

«‹›8 »

evil_bunnY: Apr 2, 2003

echo465 posted:

Double check compatibility between cards and SFP+ modules before you buy. I had planned to use all Cisco SFP+'s, but Intel x520 cards don't work with Cisco SFP+ modules. http://www.intel.com/support/network/adapter/pro100/sb/CS-030612.htm

That's good info regardless, but you don't use modules with twinax, the cables plug into sfp+ ports directly.

# ¿ Apr 14, 2012 10:11

Adbot: ADBOT LOVES YOU

# ¿ May 9, 2024 12:39

evil_bunnY: Apr 2, 2003

Docjowles posted:

Performance has been poor, but I only have 7200 RPM SATA drives in there so I don't expect it to scream. By which I mean day-to-day usage is totally fine, but something like shuffling VMDK files between LUNs takes many hours even though they are on the same drat physical device. I've never had something nicer so I don't know how typical that is.

All these devices look nice if you have no clue about storage (web UI!) but the performance tends to disappear as soon as your arrays involve parity.

And with those spindle counts it's not wonder your VMs hate you.

evil_bunnY fucked around with this message at 01:40 on Apr 17, 2012

# ¿ Apr 17, 2012 01:36

evil_bunnY: Apr 2, 2003

How do you lie your 3220's?

# ¿ Apr 17, 2012 17:07

evil_bunnY: Apr 2, 2003

Wonder_Bread posted:

I don't believe this is the case at all. It doesn't even make sense to me. The E1000 is an emulated 1Gb Intel NIC, I don't see how it could go faster.

Because nothing in the virtual hardware rate-limits it? It's easy enough to test with a ramdrive and dev/null

# ¿ Apr 18, 2012 00:02

evil_bunnY: Apr 2, 2003

Sleepstupid posted:

I want to make another "instance" of that VM, is that possible? Can I just copy the one I already converted or do I have to re-convert the same laptop again (which took over 3 days)?

Sleepstupid posted:

OK, I found the Datastore browser, copied the existing files to a new folder, created a new VM and pointed it at the new folder. Now when I try to start the new VM I get a blue-screen during windows boot. Did I miss anything?

Thanks for all the help

Open the VM settings and double check all your settings. Also if it bluescreens chances are there's a dump to be examined somewhere.

# ¿ Apr 19, 2012 20:55

evil_bunnY: Apr 2, 2003

Sylink posted:

Is the main bottleneck for running VMs the hard drive speed?

In a test env when you're rebooting/changing/copying/installing poo poo all the time you'll almost certainly be hitting I/O bottlenecks all the drat time. Even a single SSD solves that problem handily.

# ¿ Apr 20, 2012 12:15

evil_bunnY: Apr 2, 2003

Rhymenoserous posted:

There is no good reason to do this (and it isn't as easy as P2V)

Actually there are (placating support engineers who blame your running virtual for poo poo that's got nothing to do with it).

Calling back 45mn later with: "oh yeah and BTW this is on bare metal" is pretty hilarious.

# ¿ Apr 20, 2012 16:53

evil_bunnY: Apr 2, 2003

Are the host bits identical (on the vMotion source)?

# ¿ Apr 26, 2012 16:19

evil_bunnY: Apr 2, 2003

sanchez posted:

An Equallogic PS series can be had for just under 20k with some SAS drives. I think that's good value, local storage of VM's is just a pain in the rear end.

This. 12 SAS 10K drives isn't going to break the bank. And you can replicate on schedule to another one.

# ¿ Apr 30, 2012 16:53

evil_bunnY: Apr 2, 2003

three posted:

I was under the impression that 10K SAS drives are pointless to buy. They provide 33% less performance, but are not 33% cheaper.

They're only worth buying if your SAN's 7,2k drives are SATA, IMHO.

Misogynist posted:

Both of these postulates depend entirely on whether you're talking about 2.5" or 3.5" drives.

And this, and which ones your SAN can take.

CrazyLittle posted:

What's a good setup for 30-40 VMs? (simple webservers, etc)

Metrics please.

evil_bunnY fucked around with this message at 20:36 on Apr 30, 2012

# ¿ Apr 30, 2012 20:33

evil_bunnY: Apr 2, 2003

After it dumps all my data there's no way I'd trust it with it again.

# ¿ May 7, 2012 07:53

evil_bunnY: Apr 2, 2003

Why aren't you just storing the VMs on the existing storage?

# ¿ May 9, 2012 19:18

evil_bunnY: Apr 2, 2003

Erwin posted:

Also, what in the world shared storage do you have that under-performs local storage?

It take 40 grand to outperform a local $150 SSD.
And yeah svMotion would have trivialized the whole thing.

# ¿ May 10, 2012 19:08

evil_bunnY: Apr 2, 2003

How about you take a look at the load and IO sizing first then worry about the hardware.

10k simultaneous users could mean anything depending on typical transaction load per session.

# ¿ May 13, 2012 11:54

evil_bunnY: Apr 2, 2003

Free ESXi is also ram limited, so workstation's better unless you have a license.

# ¿ May 15, 2012 08:12

evil_bunnY: Apr 2, 2003

Mierdaan posted:

Right, I'm just saying that when you're figuring out how much RAM you're entitled to use, that's typically dictated by how many CPU licenses you've paid for. When you're running unlicensed, that limit has to be coming from somewhere else, and I didn't know where that was.

Also that FAQ still mentions vmware server :laugh:

Moey posted:

PM sent, not sure if I should put companies names up here.

We know where you live, buddy!

# ¿ May 15, 2012 18:52

evil_bunnY: Apr 2, 2003

Misogynist posted:

Scott Lowe was such a good blogger before he became an EMC advertisement

This. It's so annoying sifting through the bullshit now.

# ¿ May 18, 2012 14:10

evil_bunnY: Apr 2, 2003

Yes you'll want to assign as much as the VMs will need to ESXi

# ¿ May 19, 2012 00:21

evil_bunnY: Apr 2, 2003

Corvettefisher posted:

I think PCoIP is performing better too.

What the gently caress

# ¿ May 22, 2012 20:26

evil_bunnY: Apr 2, 2003

Corvettefisher posted:

Would anyone care, or would want me to post up some similar projects that I deal with at work? Just so people who don't have too much VM experience can see the things that come up day-to-day? Sorta like simlabs, but just real world scenarios of if you got a job as a vmware engineer/admin what you might see.

Do it so we can make fun of your storage planning! :laugh:

I kid, I kid. Sounds good.

In other news, the entire central virtual infrastructure at our uni just poo poo the bed. This happened 4 hours ago, and they're still not up. These are the people who got mad when I told them "thanks but no thanks, I'll roll my own" not so long ago.

Methylethylaldehyde posted:

One actual question: What do you guys do for an offsite disaster recovery site? We're looking at setting one up for business continuity in the case of fire/earthquake/volcanic implosion, and aside from setting up a DFS replication group and using DPM to Disk to Disk system state backup all our VMs, I can't think of what else we'd need to do.

15mn snapshots async-replicated to DR site, and a spare ESXi host for essential services.

Test the poo poo out of your DR plan. Don't cheat (you don't get to touch anything on the protected site once the test starts. Hope you printed your plan and passwords).

evil_bunnY fucked around with this message at 11:08 on May 25, 2012

# ¿ May 25, 2012 11:00

evil_bunnY: Apr 2, 2003

Internet Explorer posted:

Oh, maybe I misunderstood the question. I assumed he was talking about connecting to a remote virtual machine, not running something like VMware Fusion. Yes, some people using Macs do that. Most use Bootcamp, though. Less of a headache in the long run.

Fusion 8 DX performance is atrocious.

# ¿ Jun 1, 2012 23:10

evil_bunnY: Apr 2, 2003

Kachunkachunk posted:

This is definitely an area that will be developed more in ESXi. Even vmotion might be possible later if each server has the same GPU. It's just more processor and memory to allocate/reserve/share/balloon, etc.

Some of my researchers are going to wee themselves when we finally have hosts with a whole bunch of GPUs.

# ¿ Jun 2, 2012 11:43

evil_bunnY: Apr 2, 2003

Unless it's massively improved recently, no.
And last time I checked the VSA was entry level SAN money anyway.

# ¿ Jun 3, 2012 21:58

evil_bunnY: Apr 2, 2003

Put something like Nexenta on it then.

# ¿ Jun 4, 2012 15:50

evil_bunnY: Apr 2, 2003

FISHMANPET posted:

Gonna go ahead and quote this and point that while it may have been way more economical in a capital (CAPEX) sense, it's certainly not economical in an operations (OPEX) sense.

If you don't need HA it's certainly fine.

# ¿ Jun 4, 2012 17:37

evil_bunnY: Apr 2, 2003

Bob Morales posted:

Argh.

Can you bench the performance? Show him the production system, ask him what he thinks, tear him a new one when he fumbles. Adding vCPUs is super counter-intuitive because of scheduling, but he doesn't seem to have a basic understanding on IO requirements either, which is less excusable.

# ¿ Jun 6, 2012 10:54

evil_bunnY: Apr 2, 2003

Mierdaan posted:

How much of an issue is cpu scheduling now? I still hear it brought up constantly but I guess I don't know how much relaxed coscheduling has improved the situation since the old strict days where it was a big concern.

Early post on the ESX 3 scheduler:
http://communities.vmware.com/docs/DOC-4960

ESX 4:
http://vmwise.com/2010/07/09/what-is-co-scheduling-anyway/

# ¿ Jun 6, 2012 14:50

evil_bunnY: Apr 2, 2003

the spyder posted:

My sales rep just tried to sell me vSphere standard edition... For my two, 192gb ram, dual 8 core hosts... He claimed it supported HA and vMotion, and that I only needed one license...

Correct me if I am wrong, but for vSphere Essentials Plus and vCenter foundations, I am looking at $7500~ in licensing for these two hosts...

vSphere is per socket, so you need 4 licenses, and that wouldn't cover your physical RAM.

HalloKitty posted:

Does it not matter if you have cores to spare?

You'll get better performance on the vCPUs you do assign when things get busy.

DrOgdenWernstrom posted:

http://www.vmware.com/products/datacenter-virtualization/vsphere/small-business/compare-kits.html

Under vRam entitlement for Essentials Plus, why does it say "32GB (192GB total)" if it's cumulative?

It's 3 (2 sockets) hosts max on the bundles, 6 sockets.

evil_bunnY fucked around with this message at 10:15 on Jun 7, 2012

# ¿ Jun 7, 2012 10:10

evil_bunnY: Apr 2, 2003

Misogynist posted:

Also remember that this is a vRAM entitlement, not a physical memory entitlement for the host. That means you have some room to keep free RAM for HA failover and whatever else you need it for.

Or don't, if you have non-critical VMs 8(

# ¿ Jun 7, 2012 15:18

evil_bunnY: Apr 2, 2003

Put the vSphere client on your bootcamp partition and put the vm on its own desktop, and pronto!

Then ctrl+arrows to switch desktops.

# ¿ Jun 12, 2012 01:31

evil_bunnY: Apr 2, 2003

Sylink posted:

Do I really give a poo poo about the frames in a SAN that much down to what each set of bytes is doing?

It kinda makes or breaks performance, so yeah?

Sylink posted:

For instance, I can set up networks and have an intermediate understanding of how it works, but I couldn't tell you the exact structure of a packet because it almost never relates to the tasks I tend to be doing.

Until poo poo breaks in weird ways, and you don't understand how the layers interact with each other.

# ¿ Jul 2, 2012 19:18

evil_bunnY: Apr 2, 2003

Misogynist posted:

Eh, I'm torn on this one. It does, but the FC layer is at fault for so few performance issues in small SANs that you can pretty much ignore it if you're not doing anything stupid.

Now keep in mind my FC knowledge is pretty much zero compared to yours (ethernet forever) but the same kind of logic can apply to say, a data stripe once you're on the storage head.

# ¿ Jul 2, 2012 20:12

evil_bunnY: Apr 2, 2003

FISHMANPET posted:

Actually the plan has 8 10GBe to each host .

What the gently caress man. I'm running 4 and that's only because I've got fuckall else to do with my switchports.

# ¿ Jul 4, 2012 00:28

evil_bunnY: Apr 2, 2003

The VM's can only be run under Hyper-V or?

# ¿ Jul 6, 2012 02:04

evil_bunnY: Apr 2, 2003

Just the big SSD would suffice for that, if it's any good.

# ¿ Jul 16, 2012 09:39

evil_bunnY: Apr 2, 2003

Mierdaan posted:

I'm not sure who's actually buying VMware's VSA. It's a product, for sure, but that doesn't mean they're selling licenses.

Out of ~250 customers at the consultancy I used to work for, not one ran VSA. Which is a shame because it could be brilliant.

# ¿ Jul 17, 2012 22:37

evil_bunnY: Apr 2, 2003

Corvettefisher posted:

Also snapshots are really backups.

Hehehehehe

# ¿ Jul 21, 2012 23:41

evil_bunnY: Apr 2, 2003

It's got dumber scheduling than ESXi IIRC, so that would mean yes.

# ¿ Jul 25, 2012 09:34

evil_bunnY: Apr 2, 2003

If you look at the reviews against the sandy bridge stuff, they're about a 15% improvement at equals clocks if I remember right.

evil_bunnY fucked around with this message at 16:58 on Jul 27, 2012

# ¿ Jul 27, 2012 16:48

Adbot: ADBOT LOVES YOU

# ¿ May 9, 2024 12:39

evil_bunnY: Apr 2, 2003

Misogynist posted:

Are you bottlenecking on CPU speed? It's by far the least utilized resource in our environment. We average about 20-25% across the cluster.

Another little gold nugget right there. In most environments, even with TPS and all the other crap, you'll still run into IO/RAM contention rather than CPU.

# ¿ Jul 27, 2012 19:37

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > Virtualization Megathread V2: VMs inside VMs

«‹›8 »