News for hackers. Stuff that matters.
Re:Microsoft and Red Hat partner by diakritikal on Wednesday November 04, 2015 @09:58AM (#10506971)
In the last 5 years we've experimented with Linux on top of hyper-v a few times.

The basic I/O, compile & runtime performance was significantly inferior to xen & kvm (didn't bench against vmware, phasing it out due to cost), it wasn't worth any effort to even deploy apps for testing.

Therefore I don't see Linux on hyper-v being a compelling option for the cost conscientious technical officer or lead engineer.

Re:Microsoft and Red Hat partner by Someone1234 on Wednesday November 04, 2015 @11:16AM (#10507565)
With all due respect: I don't buy it. That isn't how the technology works. Compilation in particular would never even hit the hypervisor layer regardless of which one you were utilising. It is CPU/memory bound.

The difference between hypervisors for IO is tiny. They're all slower than native, but vary very little between one another. That's why the manufacturers have all almost given up trying to use performance for comparison, and instead argue value adds/management/automation/cost.

If you really found one specific hypervisor slower than another (regardless of which two), I'd want to look at your exact test conditions since that has never been the case for me (everything else being equal). And even the manufacturers don't argue that point much.

There are plenty of reasons to e.g. pick AWS over Azure, but Hyper-V performance is not one of them. That's just FUD.

Re:Microsoft and Red Hat partner by evan_miller on Wednesday November 04, 2015 @01:20PM (#10508447)
You're missing something very significant: Compilation jobs involve a lot of forking. Forking, memory accesses, and TLB differences across various VM architectures can have an enormous impact on performance.

For example, we found two orders of magnitude performance difference between PV and HVM on ec2 on fork-heavy CI jobs. A difference between 3 and 30 minutes. This difference is trivially reproducible by measuring the time taken to complete a fork() syscall. As the memory size of the parent process grows, the difference becomes more pronounced.

Here's actual data I collected from c3.large instances, using a contrived Ruby one-liner to benchmark fork():

PV: [ec2-user@PV ~]$ ruby -e'$ref=[]; (24..29).each {|n| $ref << "X" * (2 n); start = Time.now.to_f; fork.nil? and exit!; puts Time.now.to_f - start; sleep 1}'

0.00550532341003418

0.011992692947387695

0.02542281150817871

0.05213499069213867

0.10562920570373535

0.21283364295959473

HVM: [ec2-user@HVM ~]$ ruby -e'$ref=[]; (24..29).each {|n| $ref << "X" * (2 n); start = Time.now.to_f; fork.nil? and exit!; puts Time.now.to_f - start; sleep 1}'

0.0005323886871337891

0.0007219314575195312

0.0012335777282714844

0.002183198928833008

0.004210233688354492

0.008102178573608398

As you can see, the difference is considerable. This is all on Amazon ec2, the only difference between these VMs is PV versus HVM.

(edit: HN formatting seems to be eating the exponent operator in the "2 to the n" expression. It's a double *)

Re:Microsoft and Red Hat partner by merb on Wednesday November 04, 2015 @02:16PM (#10508836)
that wasn't always the case. i remember times where pv could be faster on a lot of forks. however you can't compare pv/hvm vs kvm vs hyperv since the difference between pv and hvm is caused by a totally other aspect than the comparsion between hvm, kvm and hyperv. also note hyperv got way better in windows server 2012.

also note that fork() involvs "lots of" io computations which could also be a driver problem and they evolved in newer kernels, too. and also fork() could syntactially be different on hosts with ballooning and hosts without. especially on kvm.

edit: as of today we switched to hvm instances (however we are on many many micros so that performance boost is likewise zero) on our inhouse hardware we are varying between kvm and vmware. however I dislike vmware more and more. vcenter is just unusable (flash) and consumes a shitload of memory. so mostly we use vmware api calls which makes vmware unnecessary somewhat. also we sell our software as a appliance with a xenserver host and mostly we tested our software an all ends and we are mostly disk io driven and that barely changes between them. there are some aspects that could improve performance but that depends more on how you configured your virtual disk.

Re:Microsoft and Red Hat partner by cthalupa on Wednesday November 04, 2015 @03:30PM (#10509447)
>that wasn't always the case. i remember times where pv could be faster on a lot of forks

32bit, or 64bit pre-virtualization hardware extension (vt-x/d, ept, amd equivalents) are really the only ways this should have been the case. Each time you make a fork() system call on 64bit PV you're making an extra context switch and flushing the TLB that an HVM instance does not need

Re:Microsoft and Red Hat partner by wslh on Wednesday November 04, 2015 @12:38PM (#10508166)
I don't buy it either and it is not my experience. My bad experience with hyper-v was running a desktop Linux but was expected because other factors like 3D/2D acceleration are not supported.

I would expect a full description of the benchmarks and setup to follow up.

Re:Microsoft and Red Hat partner by stronglikedan on Wednesday November 04, 2015 @11:47AM (#10507778)
I have a developer who can show a marked difference in performance between VMWare Fusion (paid) and VirtualBox (free) running Windows 7 on OSX. Same Windows ISO used to install both, and both running on the same hardware and OS, with exact VM configs (as close a you can get between the two). I'm talking a night-and-day difference - VMWare Fusion is usable, but VirtualBox is not.

After reading your comment, and admittedly not knowing much about hypervisors (yet), I am genuinely curious as to what else could be causing this drastic difference in performance, if not for the hypervisor itself. Any ideas?

Re:Microsoft and Red Hat partner by Someone1234 on Wednesday November 04, 2015 @11:53AM (#10507826)
I was more talking about actual hypervisors that run on hardware. Client hypervisors (those that run on an underlying OS) are an entirely different thing.

Real hypervisors are utilising a lot of hardware features to accomplish jailing. Client hypervisors may or may not be, depending on how they're implemented, and or how they're integrated into the underlying OS (e.g. specialised drivers, just simulating hardware features, etc).

Re:Microsoft and Red Hat partner by 5ilv3r on Wednesday November 04, 2015 @11:58AM (#10507871)
That's not a very good example since both of those are desktop applications providing virtualization and are expected to play nice with other desktop apps. You can virtualize in a web browser too, but that doesn't make it a good high performance idea.

You should be looking at dedicated virtualization hosts to do an apples to apples comparison, like vmware on bare server metal vs hyperv or any of the free tools also on metal. Not on a cpu throttled desktop.

Re:Microsoft and Red Hat partner by wslh on Wednesday November 04, 2015 @12:33PM (#10508124)
There is a big impact in performance if you are running a graphical user interface instead of a server only with the console. VMware put more efforts in the UI than other vendors.

I would always separate benchmarks for server or desktops.

Re:Microsoft and Red Hat partner by voltagex_ on Wednesday November 04, 2015 @05:01PM (#10510114)
I'd be interested in the same test with the latest VirtualBox 5 - you can use HyperV by switching a setting in the VM control panel.
Re:Microsoft and Red Hat partner by jacquesm on Thursday November 05, 2015 @12:17PM (#10514598)
It might be worth your time to run that same benchmark without any hypervisor to establish a baseline.
Re:Microsoft and Red Hat partner by aexaey on Wednesday November 04, 2015 @10:53AM (#10507389)
I'm getting approximately the same CPU & storage performance out of Azure's A0 machine w/Debian8, as from AWS t2.micro instance with spinning rust storage and Debian8 (both are cheapest offer available, and price is quite similar between them).

Network on Azure though is a totally wrecked thing:

- A0 instance is capped to a miserable 5 megabit/s uplink (downlink is fine);

- If you want TCP/UDP ports open, you need to open them one-by-one, going through a 2-page wizard with 8 fields in total to fill in for each port you are opening, and then waiting 15-20 seconds for each "change" to be applied. Good luck opening a 1024-port range for your SIP server;

- You can't open ICMP/GRE/IPSEC or any other custom protocol;

- There are ton of really weirdly configured network gear between your VM and the Internet - packet drops and private IPs in traceroute, etc...

- On a plus side, you can enable so called "direct server return" and you will get incoming packets with your public IP as a destination IP; a really nice feature sorely missing on AWS.

Re:Microsoft and Red Hat partner by gruez on Wednesday November 04, 2015 @12:12PM (#10507974)
>A0 instance is capped to a miserable 5 megabit/s uplink (downlink is fine);

I thought it was 10Mb/s?

>If you want TCP/UDP ports open, you need to open them one-by-one, going through a 2-page wizard with 8 fields in total to fill in for each port you are opening, and then waiting 15-20 seconds for each "change" to be applied. Good luck opening a 1024-port range for your SIP server;

...or use the powershell cmdlets and do it in one line

Re:Microsoft and Red Hat partner by aexaey on Wednesday November 04, 2015 @12:31PM (#10508112)
> I thought it was 10Mb/s?

Alas... Also, in my experience, throttling is really slow to react, i.e. you get a chunk of data out at 10-20Mb/s, then everything freezes for few seconds until average would drop below 5Mbps, then another speed-up/slow-down cycle. Long-term average is exactly 5Mbps as per their spec-sheet [1].

[1] http://download.microsoft.com/download/4/1/1/411621F0-D0BF-4...

[2] https://www.oaklight.us/2014/06/azure-network-speed-quick-te...

Re:Microsoft and Red Hat partner by X-Istence on Wednesday November 04, 2015 @12:46PM (#10508223)
I don't see why private IP's in a trace route is necessarily wrong...

Many large providers are numbering their hops with private IP space and routing public space across it because they don't want to waste /30's of public space for each point to point.

Yes, /31 is an option, but there were a lot of edge cases or network gear that didn't support it.

Re:Microsoft and Red Hat partner by aexaey on Wednesday November 04, 2015 @01:13PM (#10508403)
Yes, you are absolutely right that IPv4 addresses are scarce.

Which bring a related question - what's up with none of AWS/Azure/GCE supporting IPv6?

Re:Microsoft and Red Hat partner by X-Istence on Saturday November 07, 2015 @03:57PM (#10526198)
Unfortunately IPv6 in cloud deployments is still very much lacking. OpenStack is one of those where IPv6 has only recently become something that is being built into OpenStack.
Re:Microsoft and Red Hat partner by mobiplayer on Wednesday November 04, 2015 @01:20PM (#10508452)
Hi aexaey,

There are some things wrong on your comment, please let me comment on them:

I don't think A0 are capped at all, I don't know where did you get that:

user@A0VM:~$ speedtest-cli

Retrieving speedtest.net configuration...

Retrieving speedtest.net server list...

Testing from Microsoft Corporation (40.113.XX.XX)...

Selecting best server based on latency...

Hosted by KsFiberNet (Wichita, KS) [45.26 km]: 139.703 ms

Testing download speed........................................

Download: 44.48 Mbit/s

Testing upload speed..................................................

Upload: 22.54 Mbit/s

I just had to run one single test to get these results. Multi-threading will probably pump this up (let's say a multi-thread iPerf test between two VMs on two different regions, through their public IP addresses).

(Funnily enough this VM is in Dublin, hence the latency to Wichita).

You can get access to all your TCP/UDP ports by adding an instance-level IP: https://azure.microsoft.com/en-gb/documentation/articles/vir... (you can later protect it with NSGs --> https://azure.microsoft.com/en-us/documentation/articles/vir... ). What you've seen is when you create endpoints on the load balancer that points to the VMs behind it. Now in Azure Resource Manager, spinning up a VM from the new portal will actually get you an ILPIP and no load balancer by default.

About that statement about a ton of weirdly configured network gear: Well, for starters you're doing network virtualization here :) but seriously now, the platform doesn't forward ICMPs to/from Internet, so I'm curious about those private IP addresses on a traceroute (and of course the packet drops). If you're talking about hops between you and the VIP (the load balanced public IP in front of your VM), that's probably your provider, other providers and finally Microsoft's Network. As per the rest of the path, you've said that ICMP doesn't go through, so traceroute it's not going to work (tcptraceroute also needs ICMP to work :)).

Re:Microsoft and Red Hat partner by aexaey on Wednesday November 04, 2015 @02:34PM (#10509001)
ILPIP look promising. I'll try it, thanks!

Speed test though... hm... 22Mbit/s is not exactly unrestricted, plus not sure how long is the burst that "speedtest-cli" makes. In my experience A0's uplink is burstable (but unusably slow when recovering from a burst), with long-term average being bang-on 5Mbps.

I do of course see exactly one "my" private IP in traceroute, but I'm not talking about that. I meant there are microsoft's private IPs (#12-#15) and dropped packets (#16 and on).

And of course, "platform doesn't forward ICMPs" is an annoyance in itself - i can't even ping my own VM.

    # target IP anonymized, but still have enough granularity to reproduce the traceroute
    traceroute to xxxxx.cloudapp.net (191.235.128.128), 30 hops max, 60 byte packets
     1  192.168.1.1 (192.168.1.1)  0.093 ms  0.117 ms  0.063 ms
     2  193.47.232.12 (193.47.232.12)  0.785 ms  0.785 ms  0.776 ms
     3  microsoft.mix-it.net (217.29.66.112)  1.219 ms  1.233 ms  1.225 ms
     4  xe-0-1-1-0.fra-96cbe-1a.ntwk.msn.net (207.46.42.12)  8.700 ms  8.902 ms  8.430 ms
     5  * * *
     6  * * *
     7  104.44.9.142 (104.44.9.142)  42.112 ms  39.194 ms  38.987 ms
     8  * * *
     9  be-6-0.ibr01.dub30.network.microsoft.com (104.44.4.142)  39.738 ms  38.266 ms  38.929 ms
    10  * * *
    11  ae2-0.db3-96c-3b.ntwk.msn.net (204.152.141.81)  36.986 ms  36.984 ms  36.925 ms
    12  25.149.64.247 (25.149.64.247)  36.951 ms  36.775 ms  37.152 ms
    13  10.10.132.81 (10.10.132.81)  36.917 ms  37.477 ms  37.287 ms
    14  10.60.31.77 (10.60.31.77)  37.317 ms  37.699 ms  37.899 ms
    15  10.60.31.69 (10.60.31.69)  37.653 ms  37.326 ms  37.441 ms
    16  * * *
    17  * * *
    18  * * *
    19  * * *
    20  * * *
    21  * * *
    22  * * *
    23  * * *
    24  * * *
    25  * * *
    26  * * *
    27  * * *
    28  * * *
    29  * * *
    30  * * *
P.S. #12 - lol!
Re:Microsoft and Red Hat partner by toyg on Wednesday November 04, 2015 @03:25PM (#10509396)
> P.S. #12 - lol!

In view of today's news, this is not funny at all.

		netname:        UK-MOD-19850128
		descr:          UK Ministry of Defence
		country:        GB
		org:            ORG-DMoD1-RIPE
		admin-c:        MN1891-RIPE
		tech-c:         MN1891-RIPE
		status:         LEGACY
		mnt-by:         UK-MOD-MNT
		mnt-domains:    UK-MOD-MNT
		mnt-routes:     UK-MOD-MNT
		mnt-by:         RIPE-NCC-LEGACY-MNT
		created:        2005-08-23T10:27:23Z
		last-modified:  2015-07-24T14:31:16Z
		source:         RIPE # Filtered
Re:Microsoft and Red Hat partner by sivaedupuganti on Thursday November 05, 2015 @11:27AM (#10514218)
Based on iperf tests, it appears uplink on a A0 is being capped around 5Mbps. Considering I work with Azure Networking, I will find more details on this internally.

I generally work with Azure CLI (https://github.com/azure/azure-xplat-cli) and ARM templates (https://github.com/Azure/azure-quickstart-templates), helps with simpler configuration and consistent results across deployments.

Though we are aggressively working on adding more features and improving documentation, evidently there are gaps in certain areas. Please feel free to reach out to us through Twitter (@AzureSupport) or Stackoverflow (tag-Azure), for anything we can help with your deployments on Azure.

Re:Microsoft and Red Hat partner by mobiplayer on Wednesday November 04, 2015 @03:23PM (#10509377)
Yeah, the ICMP stuff could be a bit annoying :)

My comment about the traceroute was because you previously said that you've found those weird IP addresses between the VM and Internet. That's not how I actually see it, but I get your point, too.

The ones you identify as packet loss are just the ones you didn't get an ICMP back. I don't know anything about your background, but as per my experience with networking I've stopped considering traceroute as something remotely useful further than the first hop long ago, especially on environments outside my control (i.e. Internet). There are things out there like ICMP throttling and the fact that I could send you the "TTL Expired" ICMP identifying myself with any IP address. Remember the "Star Wars traceroute"? http://www.theregister.co.uk/2013/02/15/star_wars_traceroute... (doesn't seem to work for me now).

(About #12, WTF? Weird!)

Re:Microsoft and Red Hat partner by aexaey on Wednesday November 04, 2015 @04:09PM (#10509784)
Actually, in public Internet traceroutes are reasonably reliable. It's corporate networks where you would expect weird stuff like that.

> (About #12, WTF? Weird!)

There is a fairly boring explanation to that. Similar to public IPv4 exhaustion, RFC1918-style private IPv4 addresses can easily be exhausted in a big private network as well. If you have noticed, MS has actually run out of RFC1918, and uses RFC6598 (a.k.a. 100.64.0.0/10) network in Azure for their "classic" segment. 100.64.0.0/10 was not supposed to be used as RFC1918-style general-purpose private IP space, but rather as a local side of a CGN pool. But then again, if you squint a bit, well... it's still private, so will never be advertised into the Internet, so it's fair game to use it if you run out of RFC1918, right?

Some people choose to squint even further, and declare anything that is not advertised into public BGP as a fair game. I've seen a big ISP "borrowing" 44.0.0.0/8 (amateur radio AX.25) subnet to deploy video segment of their huge "triple-play" network. 33/8 (US DoD) is also popular as "borrowed" private space, 1/8 was fairly popular until it actually became real public address space handed to APNIC (oops!). Now we have an example of using 25/8 like that as well.

Re:Microsoft and Red Hat partner by X-Istence on Thursday November 05, 2015 @01:32PM (#10515200)
This is why it is unfortunate that more companies are not using IPv6 or moving towards it.

Currently working at a large MSP, and all of our stuff is still IPv4 only. Although we happen to have a ton of extra IPv4 space, so we aren't as concerned, but there is no movement towards IPv6 at all.

Re:Microsoft and Red Hat partner by ygra on Wednesday November 04, 2015 @12:14PM (#10507984)
Wouldn't this work for configuring lots of open ports: https://msdn.microsoft.com/library/azure/dn495300.aspx?

Never used the cmdlets for Azure and it's been a while that I dealt with it, but that sounds like an issue you'd only do via the web interface if you're insane.

Re:Microsoft and Red Hat partner by aexaey on Wednesday November 04, 2015 @12:23PM (#10508041)
Wouldn't work for me, as the {Get,Add}-Azure* scripts are windows/powershell-only. But then again, existence of CLI tools means there's an API. So, with a bit of luck, we might one day see a portable CLI tool (maybe boto3 [1] ported to azure API... one can dream! :-)

Thanks for pointing those out!

[1] https://github.com/boto/boto3

Re:Microsoft and Red Hat partner by ppog on Wednesday November 04, 2015 @12:50PM (#10508250)
There is a cross-platform CLI (based on Node.js), which runs on OS X and Linux as well as Windows - I don't specifically know if it has commands for opening ports though. There's information and links at https://azure.microsoft.com/en-us/documentation/articles/xpl... if you want to check it out.
Re:Microsoft and Red Hat partner by weazl on Wednesday November 04, 2015 @12:43PM (#10508212)
Try their xplat cli [1], it's built on node/npm. I'm on Windows but still prefer this one over the PowerShell one.

[1] https://www.npmjs.com/package/azure-cli

Re:Microsoft and Red Hat partner by kefka on Wednesday November 04, 2015 @11:06AM (#10507486)
But the game isn't the performance on a machine, or even a few machines.

The game are tools like Kubernetes and Apache Mesos. I have X thousand VMs spread out through the world, with a few hundred per datacenter for geo-stability.

If a network link fails, no biggie. Still runs. If a rack dies, still no biggie. Redundancy makes it work. Earthquake/tsunami/natural disaster ? Your stuff still runs, with a bit higher latency and slower throughput.

Re:Microsoft and Red Hat partner by pweissbrod on Wednesday November 04, 2015 @10:23AM (#10507161)
I find this interesting. I see many conflicting and biased studies comparing cloud/hypervisor performance.

There are a great number of variables that come into play with configuring each hypervisor. Also, considering continual updates, a solid analysis 6 months ago may be stale. Not to mention most of the studies come from the vendors, not from independent parties.

Question: does anyone know of an ongoing/reliable performance review of the major hypervisors?

Re:Microsoft and Red Hat partner by davidgerard on Wednesday November 04, 2015 @11:45AM (#10507763)
Main problem with Hyper-V: getting into the admin console with anything but IE. Yes, really.
Re:Microsoft and Red Hat partner by overgryphon on Wednesday November 04, 2015 @04:06PM (#10509770)
Vmconnect(1) is a local exe. The best experience for a console to your Hyper-V VM is remote desktop, but that requires a Windows guest and virtual networking. If you are trouble shooting networking or boot, or are using Linux guests, vmconnect is very easy to use (no network required) and works great.

I'm not sure of any way to get a console through IE. Can you explain?

(1) https://technet.microsoft.com/en-us/library/dn282274.aspx