New ESX Platform - Opinions & Configuration queries.

[spoon_]

Not at all.

It will be part of the vSwitch0, I wanted the VM to be part of the already management looking like group. Is that gonna be a problem?

[spoon_]

I have got more info from the vendors.

41 assigned CPU cores in vSphere vs 36 physical available in 3 hosts [3 boxes, dual processors, 6 cores each] - would this be ok?

Moby, how much I can realistically over allocate the processor resources?

On the RAM side of things I'm allocating 102GB out of 192 available so we are good here.

[spoon_]

2 more VMs appeared from nowhere meaning I have 43 cores to allocate and only 36 physical available.

Should I be worried performance wise?

[Splash]

I'd say that will depend largely on your workload and the degree of resilience you're aiming to build in.

[Moby-Dick]

2 more VMs appeared from nowhere meaning I have 43 cores to allocate and only 36 physical available.

Should I be worried performance wise?

on the whole no - many production environments will run just fine at a 4:1 vCpu:Core Ratio - of course if those 43 vCPU's you've specced are running a render farm or seti 24/7 then you'll have contention , but based on what you've specced , you'll be fine.

Be mindful of virtual sprawl , the number of VM's will creep up and up if you dont have the environment in control.

[spoon_]

Slight change of plan - I'm getting another QUAD NIC.

Basically the plan is to run VMs off the NFS datastore using EtherChannel and two NICs and iSCSI for the SQL databases again with two EtherChannel'd NICs.

This will also allow me to run redundant link to the DMZ switch.

NetApp shelf got installed and its up and running together with 3 other shelfs. Aggregate across all four was created to increase IOPS and general performance - this got recommended as best practice by NetApp. I would just provision the whole shelf to this project but apparently you lose performance. Having the same amount of storage spread across 4 shelf its much more efficient.

vCenter is giving me sh** when I'm trying to create Systm DSN as database is on separate clustered server, somebody should go to Microsoft and blow them up for the mess with ODBC drivers. There are two executables on 2008 R2 in two different locations with the same bloody name, one is 32 and one is 64 bit + you have to install SQL Server Native Client junk - took me a ages to figure this one out as vCentre install would not pick up the god damn DSN...

Now what I have to do is to sort the permissions out on my service account that I run the DSN on as it won't connect to the database - even with dbo [owner] rights...

Yeah, you have guessed it right - I had a bloody awful day!

[Moby-Dick]

I reckon you'd get away without the Etherchannel personally.

yeah the ODBC setup is always a bit of a PITA on 64 bit boxes

you need dbo on the VC database AND the MSDB one because vcenter creates agent tasks.

[spoon_]

Thanks for that Moby, it proved to be very helpful!

Once more question - when I was configuring default gateway for NFS/iSCSI storage it didn't allow me to leave this blank at all. Basically it was taking whatever the gateway was on my production network... I'm not sure what's the trick in here but from what I was reading you can only have one default gateway for all of your VMkernels...

Is this assigned per physical network controller or per host?

I'm thinking here of my web servers in DMZ which cannot have production network default gateway - kinda defeats the purpose of DMZ if you think about it even if there aren't routes probably...

Help?

[Moby-Dick]

I'm sure I've specified different vmkernel gateways before.

bear in mind your DMZ machines wont be directly accessing the NFS storage , its the hosts.

(btw I'm also answering your thread on VMTN , I just get points for those posts )

[badass]

I find it amazing that your employer has someone asking these kinds of questions doing this kind of documentation on a £100,000 project.
If they want consultation on this project, pay someone £950 a day for it, rather than getting a junior IT staff member to do it.

[loki]

You're fine to commit more VMs than CPUs. Scheduling will deal with that fine providing you're not running all your machines at 100%.

I've just setup two vsphere servers and it let me leave the default gateway blank on the nics that I'm using for iSCSI, as it's a seperate SAN vlan it doesn't need a default gateway as it's not routed.

What's your config going to be for the SQL servers as there are issues you can come into with those if you're using the balloon driver with SQL Server features like keep pages in memory? Also make sure you have sufficient disk I/O as that will be your killer, you can use an app like SQLIO to test if need be.

[Moby-Dick]

actually if you are using iSCSI and also using the HA feature , you may not be putting a gateway in , but I would add an additional isolation address of a key part of your storage - it'll give you the ability to engage HA should a host loose SAN connectivity.

[spoon_]

If they want consultation on this project, pay someone £950 a day for it, rather than getting a junior IT staff member to do it.

I wouldn't necessary call myself a "junior" IT staff member, well I would if you can find junior members managing your NetApp environments...

FYI, consultant has been assigned to do the work at £1000/day so no reason to question my abilities.

Plus there is nothing wrong with learning on the job, at the end of the day you have to learn one day, don't you?

Thanks.

[Moby-Dick]

so the consultant isn't actualy doing the design for a grand a day ? what exactly are they doing given your paying twice the market rate

[oolon]

I just did 100k migration, and there are two of us working on it. I think any sysadmin worth their money should be able to run a project of the 100k level. Lets assume they are on 30k when you take into account what the employer pays in tax, that less than 3 years wages. Not that much really. If it was a 1M project that you might have to ask who is best placed to make the decisions.

[badass]

Plus there is nothing wrong with learning on the job, at the end of the day you have to learn one day, don't you?

This is true. This works well for you in terms of learning but not on your organisation. Getting a junior member of staff to design an infrastructure based on software they have little experience of makes wrong decisions almost certain. The impact could be anywhere from very minor to causing the whole thing needing to be started from scratch. These simply cannot be blamed on the member of staff since the job was beyond what could ever be expected from them.