Recommended Posts

Hey Guys,

I just finished converting one of my HP MicroServers to run ESXi on a 8GB Kingston SSD with 3 7200RPM Hard Drives.

Unfortunately, regardless of which hard drive I place a VM on, the performance is terrible. It took 4+ Hours to install Windows XP and 2 Server 2003 R2 VMs simultaneously. I've used Thick Provisioning - Lazy Zeros.

Any ideas? Ideally from other people running on a MicroServer.

Thanks

8GB is more than enough for a Boot Drive. It's only a few hundred meg of Files.

All the VMs are sitting on the 7200RPM Drives.

How much memory ?

I have 7 servers running ESXi and some are ?250 1U's which run a few linux boxes but the rest run Windows 2003/2008 just fine, even on low memory.

Depends on the RAM and disk config of the server.

As you're installing the VM on a disk that's not the primary/OS disk, I can't see it being the issue.

RAID array for the 3 disks? Hardware RAID if so?

Any other VM's running at the same time? Possible IO issue?

What about trying one hard disk in the system (with the 8GB ESXi OS drive) and see can you create a VM successfully then. Add an additional disk if it works and check performance. Do the same for the 3rd disk.

Check for physical hardware issues. Try different SATA cables and/or, if possible, different hard disks.

Can't think of anything else at the moment. I take it the ESXi interface performs fine and is as snappy as it needs to be?

So you were doing 3 installs at the same time? Reading the media iso's from the same datastore (I assume) creating the disks for these installs on what datastore?

And you what don't see how there might be a bit of I/O problem with that?

edit: I have got 6 VMs on my little N40L currently -- no installs are not as zippy as you might hope for. But after the install the VMs run great! I keep meaning to get ftp working because uploading the iso's to the datastore is like watching paint dry as well ;)

In my setup I have 1 datastore on the 250GB disk. So the Iso for the install is on this disk, and then the disk I am writing too for the OS is on that same disk, etc. So this is not going to be optimal performance! And as already mentioned what else is running off that datastore drive at the same time your trying to do the installs, etc. etc.. No its not going to be a rocketship ;)

Who told you to try 3 installs at the same time?? If you going to do more than 1 of a specific OS, do it once and create a template to install your other copies from.

You tried to install the 3 VMs simultaneously? Would imagine you are maxing out the processor. Remember you only have a 1.5 GHz dual core processor in an N40L. I am doing something similar at the moment on my N36L can install a single VM in around 20 minutes. Would try doing one at a time

You tried to install the 3 VMs simultaneously? Would imagine you are maxing out the processor. Remember you only have a 1.5 GHz dual core processor in an N40L. I am doing something similar at the moment on my N36L can install a single VM in around 20 minutes. Would try doing one at a time

CPU Use shows being around 40%.

So you were doing 3 installs at the same time? Reading the media iso's from the same datastore (I assume) creating the disks for these installs on what datastore?

And you what don't see how there might be a bit of I/O problem with that?

edit: I have got 6 VMs on my little N40L currently -- no installs are not as zippy as you might hope for. But after the install the VMs run great! I keep meaning to get ftp working because uploading the iso's to the datastore is like watching paint dry as well ;)

In my setup I have 1 datastore on the 250GB disk. So the Iso for the install is on this disk, and then the disk I am writing too for the OS is on that same disk, etc. So this is not going to be optimal performance! And as already mentioned what else is running off that datastore drive at the same time your trying to do the installs, etc. etc.. No its not going to be a rocketship ;)

Who told you to try 3 installs at the same time?? If you going to do more than 1 of a specific OS, do it once and create a template to install your other copies from.

ISO(s) was on 8GB SSD

Server 2003 R2 1 (160GB 7200 Maxtor)

Server 2003 R2 2 (250GB 7200 Seagate)

Windows XP SP3 (500GB 7200 Samsung F1)

The SSD should have been able to keep up with installing on three separate drives.

That said, pulling a file over Samba from a completely independent local server only yielded speeds of 9.7MB/s (100Mb) yet should give me the full 1Gbit available to the Server.

Depends on the RAM and disk config of the server.

As you're installing the VM on a disk that's not the primary/OS disk, I can't see it being the issue.

RAID array for the 3 disks? Hardware RAID if so?

Any other VM's running at the same time? Possible IO issue?

What about trying one hard disk in the system (with the 8GB ESXi OS drive) and see can you create a VM successfully then. Add an additional disk if it works and check performance. Do the same for the 3rd disk.

Check for physical hardware issues. Try different SATA cables and/or, if possible, different hard disks.

Can't think of anything else at the moment. I take it the ESXi interface performs fine and is as snappy as it needs to be?

All disks are 100% separate (bar sharing the same Mini-SAS). No RAID in use at all here. All VMs running at the same time.

hardware or software RAID?? what are the brand and model of the drives??

No RAID. Listed Above.

So you have your 3 disks setup as 3 datastores? So you have 4 different datastores - your SSD, and then 1 on each of your disks?

As to your SSD keeping up???

Where were you seeing 9.7MBps to and from what? Doing uploads to datastore are SLOW as hell yes, I have see this!!

So I sure an the hell was not going to create a 2TB vmfs for my file server, so I created a RDM to the 2TB disk and then assigned this to my 2k8r2 essentials vm. Then did the same for the extra 2 750GB drives I had laying around. So my file server has 3TB in a drive pool (drivepool from stablebit) - raw disk access.. Even have smart info from the disks inside the vm this way! And fs is native ntfs, so if need be can pull this disk and plug it into anything and get my files off, etc.

So from my workstation to this server here are the speeds I am getting -- Not too shabby to a drive pool running on a VM, on a cheap dual 1.5GHz box ;)

post-14624-0-25695100-1334752632.jpg

I am nothing but impressed with the performance I am getting out of this little box -- for the price you can not beat it!!!

Where were you seeing 9.7MBps to and from what? Doing uploads to datastore are SLOW as hell yes, I have see this!!

Transfer was from Windows 2K8 R2 MicroServer N36L (Storage Server) to Server 2003 R2 (ESXi Virtual Machine) over Samba. I know it's not network drivers as pfSense is able to route at full gigabit speeds and thats sitting in a VM also now, this leads me to believe it's an I/O bottleneck.

"over Samba"

Since when does windows OSes run SAMBA? ;)

You mean using SMB? Samba is a linux suite that brings the SMB/CIFS protocols to linux ;) Not the actual name of the protocol used in windows file copy.

As to the speed -- and is your N36L box connected at gig? Keep in mind as well, just because you have a gig interface on the hardware of your N40L, does not mean the vms inside of it are using gig.. Its possible that pfsense is gig, and your 2k3 virtual nic is set or only seeing 100.

Here is the ram I ordered -- cheaper, and have had not issues with it.

20-148-347 MEM 4Gx2 | CRUC CT2KIT51264BA1339 R $42.99

http://www.newegg.co...N82E16820148347

Looks like the price went up a couple of bucks.

  • Like 1
Did u use any converters?? What networking gear u got??

No Converters, CAT5e

"over Samba" Since when does windows OSes run SAMBA? ;) You mean using SMB? Samba is a linux suite that brings the SMB/CIFS protocols to linux ;) Not the actual name of the protocol used in windows file copy. As to the speed -- and is your N36L box connected at gig? Keep in mind as well, just because you have a gig interface on the hardware of your N40L, does not mean the vms inside of it are using gig.. Its possible that pfsense is gig, and your 2k3 virtual nic is set or only seeing 100.

*slaps own head*

Dude you know what I meant by SMB/Samba/CIFS etc. All VMs are Gigabit.

Anyway, just to update thread, I resolved all issues, it was simply I/O being too high from multiple VMs being installed. It looks like ESXi cannot keep up with XEN / VMWare Workstation.

Glad to hear that you resolved the issue. I was expecting some issues to have resulted from a bad P2V conversion, etc..

ESXi should have been able to handle multi-vms with no problems...

Glad to hear that you resolved the issue. I was expecting some issues to have resulted from a bad P2V conversion, etc..

ESXi should have been able to handle multi-vms with no problems...

Additionally these are budget Hard Drives - I will be replacing them with decent 15,000 RPM Drives in the next few months.

To be honest I think 15k drives are kind of waste of money for this sort of setup. But hey whatever floats your boat.

And I don't think it would help with any sort of multiple installs all at the same time I/0 issue either.

Did you answer my previous question about what data stores you were pulling the iso(s) from and what datastores you were installing too.. Were they all the same data store, or common storage for the ISO's?

Also if your looking to serve up data off these drives -- I would prob RDM them to your OS your going to serve the files from vs creating a VMFS on the disk and then mapping that vmfs. Your going to get better performance with RDM vs vmfs I pretty sure.

I am serving up movies off one of my VMs - which has 3 drives in a drive pool (drivepool from stablebits) And getting 80MB+ per second both write and read to this share from my workstation. I personally don't think a 15k drive is going add much to this sort of I/0 over a 7200rpm drive.

And I don't think it would help with any sort of multiple installs all at the same time I/0 issue either.

Yeah, I think I/O is going to be fairly limited by the CPU.

Did you answer my previous question about what data stores you were pulling the iso(s) from and what datastores you were installing too.. Were they all the same data store, or common storage for the ISO's?

The ISO was stored on the 8GB DataStore, which is an SSD.

The OS was stored on the 500GB DataStore, which is a Samsung F1 7200 RPM

Also if your looking to serve up data off these drives -- I would prob RDM them to your OS your going to serve the files from vs creating a VMFS on the disk and then mapping that vmfs. Your going to get better performance with RDM vs vmfs I pretty sure.

Once I get some better hard drives, I'll be sure to go the RDM Method. Especially for the high-bandwidth VMs.

I am serving up movies off one of my VMs - which has 3 drives in a drive pool (drivepool from stablebits) And getting 80MB+ per second both write and read to this share from my workstation. I personally don't think a 15k drive is going add much to this sort of I/0 over a 7200rpm drive.

The limitation for serving up movies is I'm currently using PS3 Media Server for this, which yields fairly high CPU use, so I run this on the Storage Server as not to effect performance of the VMs

I just have to serve up the files via smb.. I don't have to transcode them on the fly or anything.. Just a simple file server, my popcorn hour plays the files in whatever format they are in.

So you were writing 3 different installs to the same datastore at the same time - and reading all from the same iso?? WTF yeah that going to be slow as hell! Why not just install once and then copy the install if you were installing the same OS?

"it would only take a number of VM's for it to choke."

No **** ;) But you can say that for any hardware - the number of VMs you can run at the same time is just going to vary. And what those VM's are doing.

What I can dell you is it runs my router, and is running my file server, and my test box w7 and my linux distro. The file server, the router and the linux box are on all the time..

It is more than capable of handling the number of VMs I would ever need to be running at the same time, and allow for many many more to be available when I need them for play or testing. Allowed me to combine my file server and my router boxes into a tiny little thing - that looking currently at the killawatt meter its connected to is drawing 55.5 watts ;) That is with 4 disks in it. Its been on for 366 hours and has drawn a total of 20.3KWH, so I pay like 12.5 cents per kwh so your looking at $2.50 to operate for 15 days. So about $5 a month to have it on running all the machines I need to run or play with.

And for the price you sure and the hell can not beat it for a home esxi solution or even a small business setup. No I wouldn't expect to run my enterprise VMs on it ;)

This topic is now closed to further replies.
  • Posts

    • Oh, I did. And it's even worse than I was hoping! Besides a lot of techno-babble jargon (yes I understand 100% of it but it's still all just techno-babble) there's 2 key points that make me super-weary about even considering testing this out. -- By default, after installation, a relay is automatically set up, so you do not need to care about that. * Non-chatmail apps use email servers as a long-term message archive while chatmail clients use email servers for ephemeral instant message relay. * Supporting the full variety of classic email setups would require considerable development and maintenance efforts, and complicate making chatmail-based messaging more resilient, reliable and fast. -- Basically, the end-user device is the 'server' (relay) so there is NO ARCHIVING whatsoever because every message is necessarily ephemeral. Great for techno-paranoia (and for illicit activities preferring no tracks to cover) but terrible for everybody else. It's also ironically contradictory to engineering principles of redundancies besides the transport layers due to the explicit absence of any persistent storage. Instead of 'classic email address' retaining multi-GB messaging archives on its server, now every device must retain 100% of those storage demands. (Email messages were originally meant to be short correspondences, not the multi-MB attachments boondoggle that now exists with unlimited spam engines flooding every potential recipient.) Any device swap or reset (or loss) makes the entire message history go bye-bye forever... lest there's an off-device auto-archival "relay" mechanism that's really a separate server that holds onto all transported messages (an email server) that utilizes 'chatmail email address' identities (like an email server) and its own persistent storage archive (like an email server). But... this solution is hoping to exist alongside real-world email address identities (based on the email server relay pathway) but simply render messages in chat thread format in an ephemeral manner (with contents being encrypted, and messages auto-expiring) ... In the end, it's a chat app/experience for the Web3/P2P-at-all-costs zealots. (I have accts on all sorts of federated web3 services so I understand the technical and non-technical alike.) For any practical users, however, it's just another service to download/install, register, cross-share id cards/qr codes, but know that there's no history/archive whatsoever (by design) so no account/message recovery whatsoever... update the device, install a bummed update patch, or dare upgrade your device... all history, poof, gone. Ya gotta start everything over again like they're a brand new person.
    • You've tried DuckDuckGo and Brave Search, now get serious with SearXNG by Paul Hill Over the last decade, it has become quite trendy to dump Google Search in favor of privacy-preserving alternatives such as DuckDuckGo, Startpage, and Brave Search. These search engines have done a very good job at highlighting dodgy practices by Google, such as adjusting search results based on what it thinks you’ll like (filter bubble) and stalking you around the web to advertise to you. While these search engines are good starting points when compared to non-private services like Google, there are still quite a few issues with them. For example, both DuckDuckGo and Brave Search require running non-free JavaScript in your web browser, which is comparable to running proprietary software on your computer, meaning you can be sure about what it’s actually doing in the background. Another issue is that these search engines are hosted on the respective companies’ servers, and you are using a service that you don’t control. Finally, DuckDuckGo, while offering privacy features, relies heavily on Microsoft’s infrastructure for its results and, in the past, has permitted Microsoft tracking scripts. If you are looking for a more private search solution than DuckDuckGo, Brave Search, and Startpage, then I recommend taking a look at SearXNG. It is a privacy-respecting metasearch engine that can be used via different public instances, which is useful for mobile users, or you can install it on your computer or server and run it locally with maximum control. Unlike Google, Bing, or Brave Search, which crawl the web and have their own search indexes, SearXNG is a metasearch engine, meaning it taps other search engines, stripping your identifying data, such as IP address, user agent, and cookies, in the process. Your search query is sent to the other search engines you enable before aggregating the results. SearXNG has deployment flexibility. If you are a casual user or a mobile user and don’t want to run SearXNG locally, you can use a public instance that is hosted by someone else. The main problem with this is that you are putting trust in the maintainer of the instance regarding stuff like logs that they may keep; good hosts should have a privacy policy explaining their policies. If you are trying to use SearXNG, you can also install the software on your device and then head to 127.0.0.1:8080 in your browser and search from there. While you don’t have to worry about a third-party admin like the public instances, search engines could ultimately block your IP address if they frown on you pulling in their search results locally. If you want to run it locally, it’s a good idea to use proxies or VPNs to hide your actual IP. You don’t have to worry about this with a public instance, as search engines never see your IP address. The main privacy benefit of using SearXNG is that it isolates your identity from the underlying engines that it’s capable of searching, such as Google and Bing. These search engines will only see requests coming from a generic server, so they can’t profile you and create a bubble filter that influences what results you see. This also ensures that your search engine doesn’t turn into an echo chamber that prevents you from reading alternative points of view. As a free software project, you are allowed to inspect SearXNG to make sure there are no negative features bundled inside. This sets it apart from the privacy search engines mentioned earlier because you can’t check their source code. As a meta search engine, you are not restricted to getting results from one source. Due to the fact that it scrapes content from other websites, your SearXNG instance will periodically get blocked from different providers, so it’s good to select a range of sources as a backup. While enabling all of the services will give you great results, this can make searching slower. I am personally happy with slower searches for the best results, but you can always check which providers are slowing down your search from the search results page and disable them to speed things up. If you want decent results quickly, enable the main search providers such as Google, Brave, DuckDuckGo, Qwant, Bing, and Yahoo. This way, you get wide coverage without the latency. On the Engines tab in Preferences, do note that there are different tabs, such as General, Images, and Videos, with their own providers that can be toggled and are not covered by "Enable all" while on the General tab, so be sure to dig into each. Just a note, if you want to enable everything, press "Enable all" in one tab, then hit save at the bottom of the page, then do the next tab, and so on. If you press "Enable all", then do that in each tab, and then save, nothing will stick. When I had just some of the search engines enabled, I searched “define nefarious” and results came back with the definition of “define” - obviously that was a sucky result. However, when I had everything enabled, it found dictionary pages for the word “nefarious” and even had an inline definition on the sidebar, which is quite nice too - that was delivered by WolframAlpha for anyone wondering! Probably the worst thing about this meta search engine is that the engines you select are saved with a cookie, so you must enable them on every new device you use SearXNG on, including if you decide to go into incognito mode with your web browser. Honestly, I would say this is the most annoying aspect, and perhaps if your browser lets you choose a separate private browsing search engine, then it would be best to use DuckDuckGo for this portion of your browsing. Another weakness of SearXNG is the random blocking of it by search providers. When you are on the results page, expand the “Response time” box, and it will show things like “Suspended: too many requests” or “access denied”. This is why it is good to enable several providers so that there is always a fallback to get results from. I won’t pretend SearXNG will be for everyone, however, if you enable all of the providers and put up with the slower response time, the results can be really amazing. Even if you don’t want to use it as your daily driver, keeping a bookmark handy that links to it is a good idea if you ever feel like doing a deep dive into a niche topic where other search engines are just failing to bring up any good result, due to the amount of sources it looks on. If you’re interested in radical user control over the software you use, installing SearXNG locally can also be a good idea, but be prepared to be temporarily blocked from sites if you trigger bot sensors without a VPN. Personally, I’ve opted to use a public instance, rather than install it myself. If you want to use it via a public instance, head over to searx.space to find a provider. Let us know in the comments if you have used SearXNG or its predecessor, Searx. What do you think about the quality of the results?
    • Dear Neowin, If it is not too much trouble, can you start using the new-ish designations for Insider Preview? "Experimental" is different than "former Dev" as it can apply to different models, eg 26H1 or 26H2 etc, right? No need to seed confusion IMHO. And, please "finally" update your graphics. OK?
    • Did you see their FAQ, its quite good. Have a look in the Advanced section. https://delta.chat/en/help
    • Just install Linux Mint that is a real blessing and many times cheaper because you can continue using your old Windows computer/laptop with the latest Linux updates.
  • Recent Achievements

    • Week One Done
      Woland13 earned a badge
      Week One Done
    • One Month Later
      Woland13 earned a badge
      One Month Later
    • One Year In
      bernmeister earned a badge
      One Year In
    • Week One Done
      Scoobystu earned a badge
      Week One Done
    • Week One Done
      tuben earned a badge
      Week One Done
  • Popular Contributors

    1. 1
      +primortal
      502
    2. 2
      +Edouard
      226
    3. 3
      PsYcHoKiLLa
      158
    4. 4
      Steven P.
      75
    5. 5
      FloatingFatMan
      71
  • Tell a friend

    Love Neowin? Tell a friend!