|
cheese-cube posted:A couple of years ago I watched a big honkin' Cisco 6509 grind to a halt because someone forgot to enable jumbos on a VLAN which was handling storage traffic. Things were fine for about a week until the weekend when big backup jobs kicked off. CPU usage on the 6K started creeping up over the course of about 12 hours until it pegged at 100% and the whole unit just started dropping packets. One of my colleagues had to physically attend the DC to console in and un-gently caress it. I had to drive 45 minutes to pull the cable out of a 6509 that lost its poo poo and took down a hospital one time. Felt great.
|
#
?
Mar 3, 2017 04:53
|
|
|
|
| # ? May 25, 2024 01:00 |
|
|
Thanks Ants posted:Everything's gone a bit quiet on the vSphere on AWS front - I assume there's a billion NDAs in place and it will launch when it's done, as opposed to the idea being canned. I've heard on the grapevine that VMware don't actually have anything to show yet, they apparently announced super early. theperminator fucked around with this message at 05:52 on Mar 3, 2017 |
|
#
?
Mar 3, 2017 05:05
|
|
|
Cisco firmware sucks and is buggy, news at 11.
|
|
#
?
Mar 3, 2017 06:15
|
|
|
cheese-cube posted:A couple of years ago I watched a big honkin' Cisco 6509 grind to a halt because someone forgot to enable jumbos on a VLAN which was handling storage traffic. Things were fine for about a week until the weekend when big backup jobs kicked off. CPU usage on the 6K started creeping up over the course of about 12 hours until it pegged at 100% and the whole unit just started dropping packets. One of my colleagues had to physically attend the DC to console in and un-gently caress it.
|
|
#
?
Mar 3, 2017 06:37
|
|
|
theperminator posted:I've heard on the grapevine that VMware don't actually have anything to show yet, they apparently announced super early. I can't be the only one totally missing the point of this. So you can use NSX and vSAN inside an environment which already provides EBS and VPC. Plus you get to pay VMware on top of the normal AWS costs, presumably. I get that all of the existing hybrid cloud stuff sucks, and more and more people are interested in "cloudbursting". Plus AWS is a different set of paradigms to learn (which is only partway applicable to Azure, GCE, or wherever else you go, though they have conceptual equivalents). But if you wanna move to  , why not actually learn those technologies instead of "leveraging your existing VMware training" or whatever marketing-speak comprises the only thing they have to say about it?But what's the point?
|
|
#
?
Mar 3, 2017 06:47
|
|
|
It's basically a way to have a hybrid cloud solution without having to screw around too much. Your on-premise and cloud compute will be manageable from the same interface with the same tools & workflows, you can live migrate vms to/from the cloud, you can easily set things up so that you can fail over to the cloud. NSX allows a lot of cool things like sharing the layer 2 domain across sites easily which means a VM failover to the cloud won't require IP reconfiguration, network ACLs that apply depending on the application traffic detected, or tags you've assigned to the VM rather than static policies you have to update. AWS products seem really geared towards web applications that you scale horizontally and load balance automatically etc, but there's a whole lot of business that doesn't work that way. Plenty of financial orgs with lovely monolithic java apps etc There's probably a lot more but from the POV where I work the above is what's interesting to us.
|
|
#
?
Mar 3, 2017 07:32
|
|
|
evol262 posted:I can't be the only one totally missing the point of this. It's to take money off people who think means "host it somewhere that isn't the corner of our office". Either that or it's so people can move their *~enterprise~* workloads which are just some services running on Windows Server, and get HA and a familiar management interface without having to change too much, or to buy some time until there's a chance to move the application towards a more cloudy mindset.
|
|
#
?
Mar 3, 2017 09:03
|
|
|
Thanks Ants posted:It's to take money off people who think I enjoy watching the face of those people when you tell them how much a lift and shift of SQL server costs. 
|
|
#
?
Mar 3, 2017 09:42
|
|
|
Vulture Culture posted:Speaking only for myself here: jumbo frames are worthless for pretty much any cases where you're routing packets over an infrastructure you don't own (i.e. anything Internet-facing that pushes large amounts of traffic). Oh sure, not for Internet-facing stuff, but BangersInMyKnickers was talking about NetApp stuff, so I was just assuming it was related to connecting VM hosts to VM datastores (probably not a completely safe assumption to make), which I assume would not be over the Internet (may also be incorrect). In my case there's just one switch connecting the hosts and datastores, so it seems like it should be hard to screw up 
|
|
#
?
Mar 3, 2017 12:05
|
|
|
Thanks Ants posted:It's to take money off people who think That seems like the gist. I just don't understand why these customers wouldn't just go for some other cloud platform which is intended for "pets", since I suspect that this won't actually provide any resiliency if an AZ goes down anyway... theperminator posted:It's basically a way to have a hybrid cloud solution without having to screw around too much. Your on-premise and cloud compute will be manageable from the same interface with the same tools & workflows, you can live migrate vms to/from the cloud, you can easily set things up so that you can fail over to the cloud. theperminator posted:NSX allows a lot of cool things like sharing the layer 2 domain across sites easily which means a VM failover to the cloud won't require IP reconfiguration, network ACLs that apply depending on the application traffic detected, or tags you've assigned to the VM rather than static policies you have to update. theperminator posted:AWS products seem really geared towards web applications that you scale horizontally and load balance automatically etc, but there's a whole lot of business that doesn't work that way. Plenty of financial orgs with lovely monolithic java apps etc Horizontal scaling is mostly there for relatively stateless applications or sites which simply can't be vertically scaled in any kind of efficient way once you make the front-page of Reddit (for example) Plus, horizontal=resilient in case an AZ or site goes down. Since AWS doesn't provide a way to say "my server broke; an admin needs to connect to the ilo and unfuck it". Reprovision the VM from an image, and attach your persistent storage volume to it so it can find the database again Obviously there are some advantages to traditional virt, and not paying for your own power/cooling/network/redundancy/hardware. But whether that's a cost-save depends on your business model and how much infrastructure you already have. None of this is anti-vmware. It's just 
|
|
#
?
Mar 3, 2017 15:24
|
|
|
Buttcoin purse posted:Apologies if this is a stupid question, but is there something wrong with jumbo frames, or is it just that you don't want to only get line speed on jumbo frames because you figure lots of your VM's I/Os won't actually be that big anyway? SAN side is fine for jumbo frames only, but the LAN side isn't going to be changing from a 1500mtu any time soon and I'm hitting 2gb/s pretty consistently these days with less than half density for the hardware. e: We're doing NFS which doesn't have hardware offload like iSCSI so doing what I can do lower CPU overhead is nice. And running vmotion traffic on the 1500mtu interfaces effectively tripled the time it takes to put a host in maintenance mode which bugged the heck out of me. BangersInMyKnickers fucked around with this message at 16:19 on Mar 3, 2017 |
|
#
?
Mar 3, 2017 16:16
|
|
|
My policy with all my hardware is to max out the frame size support (12k on my switches, 9k of vdSwitches and NetApps) and let the logical interfaces deal with the MTU scoping, so standard on the LAN side and jumbo on the SAN though if I have specific VMs doing traffic exclusively between each other on the LAN side I could bump the MTU there as well if it proves beneficial.
|
|
#
?
Mar 3, 2017 16:22
|
|
|
VMware's cloud push is more about providing consistent management tools for your compute and storage irrespective of where it lives. So you might have instances on prem, and in Azure, and in AWS, and in softlayer, or you might want to migrate between those services for arbitrage or regulatory reasons or whatever. So rather than having to figure out how to do the same thing in a bunch of different idioms depending on where it lives, you just use the VMWare tools and concepts everywhere and they handle the underlying complexity. Also, NSX on cloud is meant to provide east west security even within a subnet, which VPCs do not. And when you pick that VM up move it elsewhere it retains the same security profile. That's the pitch.
|
|
#
?
Mar 3, 2017 19:05
|
|
|
We're going to build a dev environment and we'll be moving two servers from 5.5 to 6.5 and I want to take this opportunity to replace our vCenter 5.5 servers with 6.5 appliances and my boss is worried it will destroy VMs because it happened in a Horizon upgrade at a place he used to work. My coworker thinks we should just get a new vCenter 6.5 server on a 30-day trial license.
|
|
#
?
Mar 3, 2017 19:09
|
|
|
anthonypants posted:We're going to build a dev environment and we'll be moving two servers from 5.5 to 6.5 and I want to take this opportunity to replace our vCenter 5.5 servers with 6.5 appliances and my boss is worried it will destroy VMs because it happened in a Horizon upgrade at a place he used to work. Well you can use the VCSA migration tool to upgrade directly from 5.5 windows to 6.5 VCSA. But your boss is dumb.
|
|
#
?
Mar 3, 2017 19:39
|
|
|
This covers everything publicly available about VMware on AWS. https://www.youtube.com/watch?v=_Rqv5Gg1VSk
|
|
#
?
Mar 3, 2017 20:49
|
|
|
big money big clit posted:Well you can use the VCSA migration tool to upgrade directly from 5.5 windows to 6.5 VCSA. But your boss is dumb.
|
|
#
?
Mar 3, 2017 21:24
|
|
|
evol262 posted:I mean, this is basically what every other hybrid cloud solution does (minus the migration from the cloud). I'm not seeing the value except "it's still VMware". But I'm not convinced that's a benefit when you're paying twice Because if you have a multi-cloud setup across AWS/Azure/On-Prem you can easily integrate them all and migrate workloads around, deploy workloads with the same tooling. quote:AFAIK, AWS Direct Connect already does a lot of this. Not all. And it's not that NSX isn't an interesting technology, it's that I legitimately don't understand why orgs looking at this wouldn't just also learn AWS tooling When you do micro segmentation and microservices at scale you spend an awful amount of time making firewall changes, networking changes to account for new services every week. Being able to deploy a new application and have the deployment process configure all of the networking/firewalling across different cloud providers/premises automatically is awesome. quote:Well, that's the cloud in general. You can spin up monster instances which will run lovely monolithic Java apps or c++ abominations which have been ported in the most half-assed way across 3 different UNIX variants then Linux (I also came out of finance) You can also hot-plug RAM & CPU so you can vertically scale without having to shut down your instance You can migrate the workload instead of blowing it away & starting again when amazon decide to retire the host your VM is running on. quote:Horizontal scaling is mostly there for relatively stateless applications or sites which simply can't be vertically scaled in any kind of efficient way once you make the front-page of Reddit (for example) Still, horizontal scaling only works if your workload supports it. some of us are unlucky enough to deal with horrible messes that don't scale like that. As a matter of preference I generally would prefer console access because if something breaks I want to know why. but there's not much business sense in that unless it's something that keeps happening.
|
|
#
?
Mar 3, 2017 23:49
|
|
|
Most software won't be able to use additional vCPUs for threading until the service is restarted FYI. Hotplug works if you're working with multiple processes contending for CPU but a single threaded process like a DB won't even look at it.
|
|
#
?
Mar 4, 2017 02:00
|
|
|
BangersInMyKnickers posted:Most software won't be able to use additional vCPUs for threading until the service is restarted FYI. Hotplug works if you're working with multiple processes contending for CPU but a single threaded process like a DB won't even look at it. SQL has supported hot add CPU without a reboot for years now.
|
|
#
?
Mar 4, 2017 02:59
|
|
|
theperminator posted:Because if you have a multi-cloud setup across AWS/Azure/On-Prem you can easily integrate them all and migrate workloads around, deploy workloads with the same tooling. AWS already provides tooling to import/export, just not 'live'. Doing it 'live' is not a significant engineering effort even to do it yourself, though. Is this possible? Sure, assuming you can converge memory (primarily a problem on Java/.NET app servers and database servers). Is it practical? Maybe not. Frankly, I find the classification of all virt as "cloud" to be dumb, despite the fact that I work in the "cloud" business unit developing traditional virt. AWS/Azure/GCE/Openstack are fundamentally different from RHEV/XenServer/Hyper-V/vSphere in terms of use case and best practice. theperminator posted:Direct connect is just a simple cross-connect as far as I'm aware, which means changes need to be made on your physical devices whenever you want to add a new VLAN etc, with SDN you can make network changes with API calls. I've used other SDN solutions. Not NSX, but that distinction isn't really relevant, since it's not that different (conceptually) from ACI or other SDN solutions, up to and including less full-featured ones (Neutron/OpenDaylight) and a number of overlay network solutions (Calico, Flannel, Weave). theperminator posted:When you do micro segmentation and microservices at scale you spend an awful amount of time making firewall changes, networking changes to account for new services every week. Being able to deploy a new application and have the deployment process configure all of the networking/firewalling across different cloud providers/premises automatically is awesome. theperminator posted:And with features like VMware's FT you can have a synchronously running copy of any VM running somewhere else, ready to take over if poo poo hits the fan FT was a joke last time I used it. I guess they support SMP now, but they still require/recommend a dedicated 10gb card for FT traffic. Even for critical services, we still set up multiple instances as HA VMs and let BGP or a load balancer fail over instead of using FT. Maybe it's more popular since they made some changes in 6. I know you can hot-plug resources. Basically every other hypervisor also does that, including some software. I do wonder how VMware is planning to do this. Either they've cajoled Amazon into exposing more 'core' functionality of their backend (still Xen last I knew), or you'll get billed for a second larger/smaller instance for the amount of time the migration takes.theperminator posted:Still, horizontal scaling only works if your workload supports it. some of us are unlucky enough to deal with horrible messes that don't scale like that. I also came out of finance. I know not all workloads support horizontal scaling. To which the question is: why run those on AWS at all? The biggest obstacle to deployments of private clouds is simply that people don't want to learn a new paradigm, rewrite software to match it, or change their workflow. That's not news. But it's best practice in the cloud for a lot of reasons. This whole post really still missed the gist of my basic question. Which is not "what is VMware and what are its capabilities?" or "why do people use private clouds?", but "why should anyone give a poo poo about vSphere on AWS" or "why not learn cloud paradigms"? If you want to have an off-site hosted VMware environment, there are a lot of options. Just use Rackspace, for example. What AWS/GCE/Azure/etc really offer isn't "virtualization hosting", but a bunch of tooling for managing/deploying/autoscaling anonymous VMs across geographically-separated environments, with solutions for object/block storage, tenant networking. It sounds like VMware is throwing all of that by the wayside or hiding it all, and the "killer feature" of vSphere on AWS is that it has "AWS" in the name. I haven't seen any real technical details on what VMware is going to do (and I'm not going to watch some video), but nothing you've offered here really tells me anything I don't already know. big money big clit posted:VMware's cloud push is more about providing consistent management tools for your compute and storage irrespective of where it lives. So you might have instances on prem, and in Azure, and in AWS, and in softlayer, or you might want to migrate between those services for arbitrage or regulatory reasons or whatever. So rather than having to figure out how to do the same thing in a bunch of different idioms depending on where it lives, you just use the VMWare tools and concepts everywhere and they handle the underlying complexity. Thanks, this pitch makes more sense (presented as a hybrid solution rather than "just" AWS). Are there any details you can provide about whether or not any of that 'complexity' is exposed via the AWS console/cli (or Azure's equivalent), or is all that a 'here be dragons' scenario where directly mucking with it will break everything? Will deploying an image to AWS give you the ability to attach it to an ELB or provision it as an AMI which can be autoscaled, or are these still 'pets' in the cloud?
|
|
#
?
Mar 4, 2017 03:20
|
|
|
evol262 posted:I'm not sure if you're secretly a VMware sales rep or if you just don't think I know what VMware offers. I do. I work on a competing solution with a similar feature set. Nope, not affiliated with vmware in any way. not all of us are pushing vendor agendas. quote:This whole post really still missed the gist of my basic question. Which is not "what is VMware and what are its capabilities?" or "why do people use private clouds?", but "why should anyone give a poo poo about vSphere on AWS" or "why not learn cloud paradigms"? If you want to have an off-site hosted VMware environment, there are a lot of options. Just use Rackspace, for example. Well sorry about that, I have no idea why businesses want to use VMware on AWS but they do, probably just so they can pull their dicks over using AWS, or the fact that they have so many availability zones or other products that the clients want to use.
|
|
#
?
Mar 4, 2017 16:48
|
|
|
theperminator posted:Well sorry about that, I have no idea why businesses want to use VMware on AWS but they do, probably just so they can pull their dicks over using AWS, or the fact that they have so many availability zones or other products that the clients want to use. Probably because they can keep the people they employ now to manage the stuff without needing hardware on prem ( You know Vcenter right? ), seems like a logical first step for a lot of companies. The dick waving probably also plays a part. "We do cloud, so innovative"
|
|
#
?
Mar 4, 2017 17:54
|
|
|
Mr Shiny Pants posted:The dick waving probably also plays a part. "We do cloud, so innovative" Don't forget the other side of the where CEOs are saying "We have do cloud or we will get left behind!" without knowing what cloud really means. Lots of companies are being told to spend a certain part of their budget on "the cloud" without any though into it making sense or not.
|
|
#
?
Mar 4, 2017 20:10
|
|
|
big money big clit posted:SQL has supported hot add CPU without a reboot for years now.
|
|
#
?
Mar 4, 2017 20:13
|
|
|
theperminator posted:Nope, not affiliated with vmware in any way. not all of us are pushing vendor agendas. That much is obvious, but  Even those of us who are (or have been) publicly associated with companies don't push those products in any way that I've ever seen. theperminator posted:
I'm still basically wondering how much "AWS" functionality is gonna be exposed, but I don't think I'll get an answer to that yet...
|
|
#
?
Mar 4, 2017 21:32
|
|
|
evol262 posted:Thanks, this pitch makes more sense (presented as a hybrid solution rather than "just" AWS). Are there any details you can provide about whether or not any of that 'complexity' is exposed via the AWS console/cli (or Azure's equivalent), or is all that a 'here be dragons' scenario where directly mucking with it will break everything? Will deploying an image to AWS give you the ability to attach it to an ELB or provision it as an AMI which can be autoscaled, or are these still 'pets' in the cloud? I don't think anyone outside of VMware really knows. The demo at VMworld last year didn't elucidate much of anything. It was just some guy clicking buttons in a browser based management console. There seem to be a number of parallel efforts at VMware to latch onto the enormous growth of public cloud providers and AWS on VMware is just one of them. Who knows what sort of actual product will come out of it though.
|
|
#
?
Mar 5, 2017 00:44
|
|
|
anthonypants posted:Turns out we still have a host on 5.1, so we're probably going to do the second thing instead of upgrading that one host IT IS ONLY HOSTING ONE VIRTUAL MACHINE, and I will continue to curse their names.
|
|
#
?
Mar 6, 2017 21:06
|
|
|
anthonypants posted:Okay I finally convinced my boss that we're going to upgrade vCenter and that host, but he thinks we should still have separate and distinct vCenter database servers instead of just using the loving VCSA That's stupid. Why?
|
|
#
?
Mar 6, 2017 21:07
|
|
|
Additional dependencies that increase the risk of failure are a feature.
|
|
#
?
Mar 6, 2017 21:13
|
|
|
big money big clit posted:That's stupid. Why? BangersInMyKnickers posted:Additional dependencies that increase the risk of failure are a feature.
|
|
#
?
Mar 6, 2017 21:39
|
|
|
You can't use an external DB with the VCenter appliance starting in 6.5. And the 6.5 appliance supports VCenter HA while the windows version does not. The windows VCenter option will probably go away at some point, so the VCSA with internal DB will be the only option. Might as well get there now.
|
|
#
?
Mar 6, 2017 21:43
|
|
|
big money big clit posted:You can't use an external DB with the VCenter appliance starting in 6.5. And the 6.5 appliance supports VCenter HA while the windows version does not. The windows VCenter option will probably go away at some point, so the VCSA with internal DB will be the only option. Might as well get there now.
|
|
#
?
Mar 6, 2017 22:00
|
|
|
Linked mode: Yes. HA is included as part of standard licensing and does not require additional licensing for the passive and witness nodes.
|
|
#
?
Mar 6, 2017 22:18
|
|
|
One of my vendors set up HA on a deliverable project they are working on for me. Which is great, because I didn't specifically request it but they took the initiative. Buuuuut something got messed up at some point and they deleted the passive and witness nodes without removing the HA config from the active node. Noticed some weirdness with certain performance stats not logging correctly, went to reboot VCSA, and it never came back up because it went in to passive mode with no active LAN interface and was assuming the witness handled failover to the other node that no longer existed. They ended up having to do a rebuild and seize the cluster hosts.
|
|
#
?
Mar 6, 2017 22:21
|
|
|
We had a ton of vcsa problems on 6.0 with xendesktop. I hope it's better by the time you have to use the appliance
|
|
#
?
Mar 6, 2017 23:23
|
|
|
adorai posted:We had a ton of vcsa problems on 6.0 with xendesktop. I hope it's better by the time you have to use the appliance That's weird, no issues with VCSA on 5.5 and XenDesktop 7.6.
|
|
#
?
Mar 7, 2017 00:14
|
|
|
vCSA has been getting more and more strict on certificate trust over the past few releases, if that helps you get started with troubleshooting.
|
|
#
?
Mar 7, 2017 00:29
|
|
|
Thanks Ants posted:vCSA has been getting more and more strict on certificate trust over the past few releases, if that helps you get started with troubleshooting. 
|
|
#
?
Mar 7, 2017 01:05
|
|
|
|
| # ? May 25, 2024 01:00 |
|
|
Internet Explorer posted:That's weird, no issues with VCSA on 5.5 and XenDesktop 7.6. How many desktops? Just curious. We only have 500 or so, and it would just stop responding occasionally during our busy logon or logoff times. VCSA is great on both of our server clusters, and the windows app is great on XenDesktop.
|
|
#
?
Mar 7, 2017 01:35
|
|