Best posts made by Francesco Provino

@ghani said in Upgrading from ESXi 5.5 to ESXi 6.5 . what are the steps need to follow ?:

@francesco-provino said in Upgrading from ESXi 5.5 to ESXi 6.5 . what are the steps need to follow ?:

vSphere upgrade path on vmware si

i searched, but that they do not provide sorted VMware upgradation pre-requistics in esxi host level, vcenter level, VM level, storage level, network level ???

Of course they require SOME of that: https://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2147289 .

More general compatibility matrix: https://www.vmware.com/resources/compatibility/sim/interop_matrix.php .

I run two VMware Horizon environments for two companies. My advice is… avoid VDI. At any cost.
The only scenario in which it makes sense is a security-tight environment (banks, military, government, strong NDA etc.).
For ANY other case, just use proper management tools, desktop imaging etc.

The VDI is a land full of pain, believe me.

Hi everyone, I was just thinking about merging two /24 subnet that are like X.X.0.0 and X.X.120.0 (running out of IPs)… some apps that run on that contains weird static-ip configuration (like SAP B1 services with hardcoded IP, CCTV ecc), so I cannot easily reconfigure everything from scratch with DHCP in a bigger subnet.

So, I was thinking about just use X.X.0.0/16 and slowly migrate the static stuff to the new mask (and migrate to DHCP with reservation, of course), but I've read too much good-old advices about "too big domain", "too much broadcast traffic" ecc to use the /16 lightly.

Do you know about any real issue about using /16 in a LAN? Here it's all modern ubiquiti switches, AP and router. The other machines are fairly new, too.

@scottalanmiller said in Any reason to avoid /16 in 2017?:

@francesco-provino said in Any reason to avoid /16 in 2017?:

@travisdh1 said in Any reason to avoid /16 in 2017?:

@francesco-provino said in Any reason to avoid /16 in 2017?:

@scottalanmiller said in Any reason to avoid /16 in 2017?:

@wrx7m said in Any reason to avoid /16 in 2017?:

A /16 seems pretty drastic when coming from a /24

Yeah, /16 is too large to actually use. Commonly /23 and /22 are used, they are no problem. In reality, /21 is perfectly fine. Even a /20 is pretty good. But once you start getting into the /19 and larger, you are just getting to some impractically large scales. A /16 is 16,000% larger than a /20, which is generally considered the largest that you can practically use.

I know this is the common sense, but… what will be the issue? I will have just 300-350 allocated IP, it's just a matter of convenience to include both the X.X.0.0 and the X.X.120.0 range in one big subnet.

I know it doesn't matter much, but the AWS VPC subnet is /16 by default :D.

After a certain point, broadcasts overwhelm actual network traffic. That's really the only thing I know that limits the size of a single network.

Ok, but I think the broadcast traffic depends only on the number of hosts in the subnet. I wouldn't put more than ~500 active IPs in this subnet, ever.

Then why go to something so absurdly large instead of just something 2-4x larger than your maximum possible usage?

Again, because there are two production networks that I want to merge together, one is X.X.0.0/24 and the other is X.X.120.0/24. It's hard to rebuild every static-ip-bounded configuration in our small maintainance window, so I plan to change it piece by piece to DHCP reservation, but I cannot do it at one time.

But, really, what's the problem with the "wasted" space? Is there something intrinsically dangerous or heavy to compute with a /16 network for modern equipment?

@kuyaz said in Dell R720 Display problem with Fedora 26 server:

1. Setup RAID-1 via PERC for 2 x 4TB HDD
2. Install Fedora 26 Server from USB Boot
3. LVM only 20GB out of 4TB for Fedora VM Host (Xen/KVM)
4. Partition :
   1. root (LVM with xfs) 17GB
   2. boot (LVM with xfs) 1GB,
   3. swap (LVM with swap) 2GB

Please correct if my setup is fine. still learning

Should I partition 20GB and leave the free space for VM guest LVM? or I do 4TB during the install, and put VM host as image file?
CMIIW.

Thanks.

You have to choose between storing the VM in LVM partition directly or in files. LVM is very solid and a little faster. The file backend is also solid today, and a lot more flexible; I suggest to use XFS as a base filesystem.
PV the whole disk array anyway.

Remeber, both thin LVM and non-preallocated QCOW files will slow the write operations A LOT.

@kuyaz said in Fedora 26 + XEN + EFI = multiboot error:

ok will give it a shot after i give up with fedora
maybe will try centos also as i am abit more familiar with centos. I also feel there are more online reference for centos compared to fedora. maybe it is just me

Fedora / CentOS removed part of the Xen compatibility in favour of KVM years ago. Of course you can still do a Xen host with CentOS, but it could be harder than with SUSE or Ubuntu.

@scottalanmiller said in XenServer 7.3 Release:

@francesco-provino said in XenServer 7.3 Release:

It's extremely reliable, powerful, has an incredible ecosystem.
That 600€ of essential license are blessed.

Except... at that price it's ridiculous. Maybe in Europe that's seen as a deal. In the US, that's pure insanity. SMBs can't throw around that kind of money. And it doesn't come with ANY support, and it isn't powerful at that price, it's actually quite pathetic. For half that cost, you can hire someone to install something more powerful.

Underpowered, insanely expensive.... where's the selling point relative to the market?

Where can you hire a sysadmin to install a KVM environment in ONE DAY that has centralized web-based management with advanced automation tools, backup API, stateless installation that fit an usb (no need for endurance on a BOSS card / additional stoarge array) and easy to manage networking across vlans (openvswitch/Linux bridges, I've been there) for 300€/$?

The performance of a single storage array is limited by the width of the PCIe lane. The only way to overcome this limitation is striping arrays across multiple PCIe interfaces.

I don't think you need something like that in a scale-up setup, we are talking about many Gbyte/s and several millions of IOPS.

@ccwtech said in RAID on SSD's:

@francesco-provino said in RAID on SSD's:

You don't "need" a separate controller, simply you will saturate both a separate SAS controller (RAID HW) and an integrated SATA one (SW RAID). Essentially, you can saturate the band of a PCIe 3.1 8x link.

So what are options when do you need to have more than 6-7 SSD's in a server then?

No problem with both software raid or hardware raid card: a modern LSI/AVAGO/Broadcom controller can take up to 255 SAS/SATA SSD in a single array. Just, don't forget that the controller will be the performance bottleneck of the array.

@stacksofplates said in Back to Active Directory, Route 53 DNS:

However I do see a big plus. If you’re using something like ZeroTier now all of your mobile devices can resolve DNS names, since you can’t control the phones DNS on cellular.

That’s exactly what we do now.

@scottalanmiller I manage an environment with SAP B1, but I don't know the answer. It used to be a very bad, clunky and slow software, now is getting a little better.

Another vote for optiplex micro, very nice built and quiet machines.

@tim_g for less than 100Tb, you could still use a traditional file server and a CBT backup like Veeam on VMware. Over 100Tb, you should really look into object storage.

@hobbit666 said in SAN vs vSAN in a brownfield environment:

@francesco-provino Why are they looking at changing?

Because of a licensing problem that took the ERP down after a CPU failure in a host. The license of a plugin was binded to a physical CPU. We are migrating away from this software in months, so it should not be an issue anymore.

@networknerd said in SAN vs vSAN in a brownfield environment:

Be careful of trying brownfield VMware vSAN. The vSAN HCL is something you must adhere to when choosing hardware to use to set yourself up for success (i.e. controller is super important - look at the HBA330 instead of a PERC, need to choose the proper drives for cache and capacity, etc.).

So if you do roll with VMware vSAN or use Starwind, I would still look at a DR cluster. You can do a 2-node vSAN cluster but need a witness running somewhere else.

Thanks, this is a very useful observation.

@coliver said in SAN vs vSAN in a brownfield environment:

@scottalanmiller said in SAN vs vSAN in a brownfield environment:

@francesco-provino said in SAN vs vSAN in a brownfield environment:

@hobbit666 said in SAN vs vSAN in a brownfield environment:

@francesco-provino Why are they looking at changing?

Because of a licensing problem that took the ERP down after a CPU failure in a host. The license of a plugin was binded to a physical CPU. We are migrating away from this software in months, so it should not be an issue anymore.

How did that lead to this?

How would this new system resolve this issue either way? Wouldn't it still be CPU locked to a single host?

Exactly what I tried to point out.

I'm in the EU, AFAIK it's perfectly legal. @matteo-nunziati , what do you think about it?

I know of many SMB that has purchased licenses like that for years, zero issues.

I have get rid of old 10/100 switches put on the floor just to make new ethernet ports available for two rooms of a small office with 3 desktop per room... two strategies:

• complete recabling of the room;
• just an AC AP for the desktops (the ports are enugh for VoIP deskphones/printers).

The thing is, recabling the rooms is hard because the cable tubes inside the walls are full so new cables will need new canalization etc... the WiFi can easily cost less (just the price of the AP + 1-2 AC pci adapters for the desktop that aren't WiFi enabled) and can reduce the cable clutting.
What do you think about it? I know cable is general more cost effective and reliable (and faster, of course), but this is a brownfield environment and one of the room will also need a new wall rack for the room switch, new patch panel...
Of course I will put one AP (ubiquiti lite) per room, and every AP will serve no more than 3 desktop + 3 mobile phones.

Hi everybody, I run out of ports in a room (solid cat6 from the main rack) , so I have to add some more...
I have at least two different way to add them:

• run 6-10 new cat6 cables for 65 meters each from the main rack to the room;
• buy a new little wall rack (60€) for the room, run two cables (fiber or copper?) from the main rack to the new little rack of the room one and connect old and new room ports using the "old" cat6 pulled from the tubes (it's only 4 years old, perfect status).

I think that the solution with the satellite rack would be more clean (there are already too many cables in the main rack with servers etc) and future-proof, and the cost would be similar: 300m of cat6 costs ~90€ vs 40€ for pre-terminated 4-strands SMF + 60€ for the little rack.
I also think that running the backbone between racks in SMF would be much more future proof (either for bandwith and diameter) than another pair of copper... what do you think about it?