@tim_g the interface is much nicer. But, I cannot upload the ISO in the datastore in ANY way. I haven't control over my virsh anymore. To get the CentOS accepted I had to downgrade the cluster level and re-up it after, other that disable completely firewalld and change the cluster default firewalld/iptables config.
Posts made by Francesco Provino
-
RE: oVirt Testing
-
RE: oVirt Testing
I'm testing oVirt 4.2 beta on a CentOS 7.4, running on an x3300m4 a lot of CPU and ram...what can I say, it's much more complex and much less useful than the standard cli-based toolstack. I've spent almost 8 hours trying make everything works. Now I cannot upload any ISO to the ISO domain, so I can only import templates from the default OpenStack Glance repo, and some of them are broken... or, maybe is the cloud-init implementation?
I hope they will fix it in the stable 4.2. The interface is nice, but for now I stick with plain KVM. -
RE: RAID on SSD's
@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.
-
RE: RAID on SSD's
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.
-
RE: RAID on SSD's
@ccwtech said in RAID on SSD's:
If you have more than 6 or 7 SSD's you need a separate controller then?
RAID 5 isn't something I have done for years...
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.
-
RE: RAID on SSD's
@ccwtech said in RAID on SSD's:
I set up my servers with RAID 10 and 10 or 15 K RPM Drives and a dedicated RAID card and an SSD Drive for CacheCade.
However with the decrease in pricing for SSD drives, I would like to start using them more in servers.
- Still do RAID 10 with SSD's?
- Is a dedicated RAID card still needed if using SSD's?
- Usually RAID 5 makes sense. Still RAID 10 if you want maximum performance. Be aware that you can saturate almost any SATA/SAS controller with the performances of 6-7 SSD.
- Apply the same consideration for the HDD.
-
RE: XenServer 7.3 Release
Maybe the reality in UE is different than in US, but that's what I see almost everyday. Or maybe it's just confirmation bias.
-
RE: XenServer 7.3 Release
Maybe in your dream world every SMB has NO VLAN, NO application like SAP that are only certified to run on VMware, no shared storage (redundant SAN/Switches, no IPOD)... Linux DM-MPIO + Clustered GFS is MUCH harder than the simple "Add VMFS" of VMware. Generally speaking, the whole clustering stack of Linux is really a mess, I have two books on that corosync/pacemaker/blabla hell.
-
RE: XenServer 7.3 Release
@scottalanmiller said in XenServer 7.3 Release:
@francesco-provino said in XenServer 7.3 Release:
@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€/$?
None of that matters. What does matter is that for 300€ you can get a better system than Vmware.
Web based management is nice, but not a deal breaker for a small shop, VMware only got that recently anyway. Backup API isn't really good for the SMB, in fact, it might be a bad thing. Stateless installation on USB, what value is there to that? Sure, use it if you have it, but really, who cares... no SMB, that's for sure. And VLANs... those don't belong in an SMB generally.
This stuff sound great, but actually, for half the price, you can probably get a better KVM deployment that isn't chock full of things you shouldn't have but might have deployed only because VMware was so expensive that you felt like you had to.
Plus, if you get VMware for 600€ and can't install KVM yourself, but need all those features... you now have to hire even more expensive support. VMware support will likely cost more than KVM support because "they see you coming."
I've fried SEVERAL USB thumbs with XS. And in small installation you usually don't have a separate array just for the hypervisor or stuff like boot from SAN.
I think any ML user (even ones that aren't familiar with VMware) are able to correctly configure a virtual switch with 10 minutes of messing with the gui or powershell.Now, try that with a standard KVM (CentOS/Fedora) installation.
Maybe it's just me, but the whole MANUAL taking up the bridge / add the NIC / activate the interface / delay the autostart of the VM since everything is up... is not the best part of life, really. No, Linux don't bring up everything in the right order, you have to script it manually and hope that no upgrade broke the compatibility with your script. And yes, you have to integrate it as a systemd unit. There are also the log to configure, of course. I'm able to do that, ok, but it takes months of my time to debug any of those little issue before releasing the whole mess in production.I don't see how backup API can be a downside...
-
RE: XenServer 7.3 Release
@scottalanmiller said in XenServer 7.3 Release:
@francesco-provino said in XenServer 7.3 Release:
@scottalanmiller and don't try to sell me the HyperV stuff, because as you stated some months ago, Windows and it's license is ultimately needed to effectively manage HyperV, so the price is not that far away from VMware :D.
That would only be a valid concern if you hadn't already stated that the situation was for people without Linux experience. If you don't have Windows, and you don't have Linux, what exactly do you need to virtualize?
Maybe I need to be even more precise: a GOOD KVM implementation requires not only a very good Linux experience but also a broad experience with KVM itself. It's easy to mess up things when you have to do almost everything in a "handmade" way. Remember, I was a fan of KVM before you even consider it for production workloads...
-
RE: XenServer 7.3 Release
@scottalanmiller and don't try to sell me the HyperV stuff, because as you stated some months ago, Windows and it's license is ultimately needed to effectively manage HyperV, so the price is not that far away from VMware :D.
-
RE: XenServer 7.3 Release
@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€/$?
-
RE: XenServer 7.3 Release
I see a big coming back of VMware in the SMB, maybe from deluded cloud customers (lift and shift burned people, mainly).
It's IMHO the obvious choice for non-Linux people.
It's extremely reliable, powerful, has an incredible ecosystem.
That 600€ of essential license are blessed.
I'm a Linux guy and I've many KVM hosts in home lab and in the field, but the VMware GUI/CLI are doing very well in the latest release. And there is also the "Veeam factor": 2/3 of KVM threads are about "how to backup KVM guest".Yes, I've always been again "that closed shit", but VMware has a very nice products that is starting to be usable again.
-
RE: Building a Veeam backup target
@tim_g said in Building a Veeam backup target:
@francesco-provino said in Building a Veeam backup target:
So, you are pointing towards the external jbod solution? The Dell chassis seems a bit dated to me...
It can't be any more dated than the MD1000. (though I didn't look up your server)
Do you want the possibility to expand (MD1000) versus having to replace 8 drives later?
I found some posts talking about MD1000 being picky with drives, don’t accept >4Tb… what do you think about that?
-
RE: Building a Veeam backup target
@tim_g said in Building a Veeam backup target:
@francesco-provino said in Building a Veeam backup target:
I'm in Italy
Ahh, I see. That changes availability and costs of things.
I use disctech.com. I know they ship internationally, but that may make them more expensive than somewhere else in Italy, or it may mean using the internal 2.5" bays are your best option.
Scott may have a lot more to add on international server hardware and related items availability or options.
Thank you @Tim_G ! I think I'll go with RAID 6 on the JBOD.
-
RE: Building a Veeam backup target
@tim_g said in Building a Veeam backup target:
It can't be any more dated than
I see your point and I agree with you, but I haven't found al that MD1000 chassis in the usual place (ebay &co), maybe the problem is that I'm in Italy?
Any alternatives? -
RE: Building a Veeam backup target
So, you are pointing towards the external jbod solution? The Dell chassis seems a bit dated to me...
-
RE: Building a Veeam backup target
@tim_g said in Building a Veeam backup target:
@francesco-provino said in Building a Veeam backup target:
I have a spare x3550m4 (64Gb of ram, single CPU, no raid controller), and I want to build a backup repository for our growing Veeam backups. I’m thinking about getting some Broadcom raid controller (any ideam of what can be the best fit for ~20Tb?), and I’m torn between using the 8 internal 2.5 slot or buying also an external JBOD, but I cannot find enclosure of my taste… any suggestion?
Using the internal 2.5" bays are out of the question, as it's hard to get a solid ~20TB RAID of that few 2.5" drives that are at lest somewhat reliable.
If it's just backups, I've so far had a lot of success with an old MD1000 with 8x 8TB HGST drives (RAID10). Even with the 3Gbps backplane limitation, I'm still able to do a full back up of about 12 TB in 14 hours. It's quite fast. If you need more speed, you can use more lower-capacity drives to hit your mark.
AFAIK 2.5 drives have the same reliability as the 3.5 ones. There are 5Tb 2.5 spindles available, so I can get 20 Tb out of 8x5Tb in raid 10... but, of course, the JBOD alternative is much more scalable (and pricier).
-
RE: Building a Veeam backup target
@scottalanmiller said in Building a Veeam backup target:
Broadcom RAID controller? What models are you looking at? I'm not familiar with their line.
Broadcom bought LSI and Avago: https://www.broadcom.com/products/storage/raid-controllers/ .
I was thinking about something like the 9270-8i for internal disks and the 9286-8e for external ones. The first is in the ~100€ range, the second circa 200€.
-
RE: Building a Veeam backup target
@fateknollogee yeah, I've already read half of the articles, very nice stuff. He is also italian like me!
But more specifically I was searching some suggestion about the design of the storage, local VS DAS/JBOD. I really thing a generic enterprise server loaded with disks is the way to go, but I still have the aforementioned doubts...