February 2015 - luv-main

Virtualization - network latency
by Peter Ross 07 Feb '15

07 Feb '15

Hi, the SAGE consultant was noticing the network latency on the "internal switch". He did not like VirtualBox + FreeBSD for that matter (I don't blame him - it was the first "get something up" quickly" configuration). Here the values: --- SageApp.vv.fda ping statistics --- 10 packets transmitted, 10 packets received, 0.0% packet loss round-trip min/avg/max/stddev = 0.308/0.635/1.364/0.319 ms It did not change much when moving to another network card (following David Lutz' recommendation: https://dlutzy.wordpress.com/2012/04/23/performance-problem-with-virtualbox/ ) However, there is a Windows box on the Oracle VM Server running, and here I have: --- SageWindows.vv.fda ping statistics --- 10 packets transmitted, 10 received, 0% packet loss, time 9002ms rtt min/avg/max/mdev = 0.554/0.605/0.646/0.037 ms So the average is not much different but the variance higher. However, he says, he sees typical 0.15 ms values on his PC with VMware Workstation. I wonder whether someone can share thoughts and experience about it (and other performance values). I found this interesting too: https://www.duo.uio.no/handle/10852/34900 I will take some time to play with various setups in the near future. Regards Peter

3 2

SSD smartctl attributes for wear indicators (was: mine-is-bigger-than-yours maildir competition)
by Toby Corkindale 07 Feb '15

07 Feb '15

I'm curious to know what your SSD wear indicators look like, from long-running Linux machines, and how long it looks like they'll last based on existing usage. You can query these with smartctl (if your drive db is too old, run sudo update-smart-drivedb first) I'll go first. These are just private machines, albeit ones doing reasonable work. Perhaps at some point in the future I'll be able to report on long-term results of enterprise SSDs, but I can't right now. Machine one: Power_On_Hours 4522 (188 days) Total_NAND_Writes_GiB 18846 Maximum_Erase_Cycle 199 Avg_Write_Erase_Ct 74 Total_Bad_Block 201 Perc_Avail_Resrvd_Space 100 This machine has been running for over 188 days non-stop, has logged nearly 19 TB of writes, and is about 2.5% of the way through it's expected minimum lifespan[1]. Estimated total lifespan time: 20.6 years. Machine two: Power_On_Hours 18326 (763 days) Used_Rsvd_Blk_Cnt_Tot 0 Wear_Leveling_Count 13 Total_LBAs_Written 2066747494 (ie. about 1080 GiB [2]) This has been running for 763 days non-stop. Like the first machine, it hasn't used any of the reserved blocks yet. It's about 1.3% of the way through its min expected lifespan.[3] Estimated total lifespan time: 160 years. -Toby 1: ie. 3000 write/erase cycles for MLC; in practice you seem to get quite a bit more though, according to testers. 2: This drive doesn't report the actual NAND writes, just LBAs written, but you can roughly convert those out; call each LBA 512 bytes, and then multiply the total by a conservative 1.1 to allow for write-amplification; we come up with about 1080 gigabytes. 3: This machine is running a cheaper type of TLC-based SSD, so theoretical amount of erase/write cycles are just 1000.

7 13

Re: Virtualization - network latency
by Peter Ross 07 Feb '15

07 Feb '15

Thanks Toby. From: "Toby Corkindale" <toby(a)dryft.net> > Here are the results of KVM and LXC based virtual machines on Ubuntu > 14.04.. > Both look a fair bit better than the results you're seeing. > > --- LXC ping from VM host --- > 10 packets transmitted, 10 received, 0% packet loss, time 9000ms > rtt min/avg/max/mdev = 0.042/0.050/0.074/0.009 ms That's in the same kernel so that should be better. Here are FreeBSD jail version 1 (shared network stack): 10 packets transmitted, 10 packets received, 0.0% packet loss round-trip min/avg/max/stddev = 0.016/0.019/0.025/0.003 ms .. and jail version 2 (VIMAGE/VNET, own network stack) 10 packets transmitted, 10 packets received, 0.0% packet loss round-trip min/avg/max/stddev = 0.236/0.280/0.321/0.030 ms results. Looks as there is significant overhead in the later. > --- KVM ping from VM host --- > 10 packets transmitted, 10 received, 0% packet loss, time 9002ms > rtt min/avg/max/mdev = 0.188/0.306/0.402/0.057 ms Compared to Oracle VM(Xen) >> 10 packets transmitted, 10 received, 0% packet loss, time 9002ms >> rtt min/avg/max/mdev = 0.554/0.605/0.646/0.037 ms It's nearly doubling the speed.. I wonder whether this is typical for KVM vs. Xen, and how VMWare ESXi is faring. I do not have a KVM or ESXi setup ready but will try. Thanks Peter

3 2

Re: Fwd: Virtualization - network latency
by Peter Ross 05 Feb '15

05 Feb '15

From: "Avi Miller" <avi.miller(a)gmail.com> > Hi, > >> On 5 Feb 2015, at 2:14 pm, Peter Ross <Petros.Listig(a)fdrive.com.au> >> wrote: >> >> Compared to Oracle VM(Xen) >> >>>> 10 packets transmitted, 10 received, 0% packet loss, time 9002ms >>>> rtt min/avg/max/mdev = 0.554/0.605/0.646/0.037 ms >> >> It's nearly doubling the speed.. > > Can you provide more information about the OVM setup? What guest type? > What virtualisation model, i.e. HVM/PVM? Here a few bits from the vm.cfg for the Windows VM on Oracle VM (no Linux VM on it yet): guest_os_type = 'windows' builder = 'hvm' vcpus = 4 memory = 4096 OVM_os_type = 'Microsoft Windows Server 2012' OVM_domain_type = 'xen_hvm' Inside is Windows Server 2012 R2. Regards Peter

2 1

Re: ISCSI export of local storage on virtualization servers
by Peter Ross 05 Feb '15

05 Feb '15

From: "Avi Miller" <avi.miller(a)gmail.com> > To: "Peter Ross" <Petros.Listig(a)fdrive.com.au> > >> From: "Avi Miller" <avi.miller(a)gmail.com> >>> Technically: yes. However, we strongly recommend you don't. However, if >>> you absolutely really want to, I recommend using NFS instead of iSCSI >>> to share the unused space. >> >> Curious question: Why do you recommend this instead of iSCSI? > > Our testing shows that NFS is less likely to cause CPU starvation of the > guests. > > Well, you can't use an exported local file system (either via iSCSI or > NFS) for any of that either. The only way you get failover is via shared > storage. Yes. The idea was to use the local disk space for other (not VM related) services. >> The exercise I am going through, satisfying proprietary software, is >> exactly a demonstration to show the benefits of open software because >> the >> later gives you the freedom of choice:-) > > Oracle VM Server is Open Source. :) Yes, more the top to bottom. Sage X3 limits DB support to Microsoft SQL Server or Oracle DB and print support to Crystal Reports and so MS Windows Server, Oracle DB with support limits my OS and virtualization choices etc. If you have a local reseller who panics about everything they did not do before (means everything that's not from Redmond) .. then it's bound to become great;-) > Have fun with your project though. I went through a similar exercise a few > years ago. Thanks, I certainly have. It keeps me busy:-) I have the infrastructure up and running but I am not really happy about my server layout, available and usable disk space, backup, mirroring, performance and administration as the main concerns (this includes more than just than SAGE). At the moment my brain is redesigning every ten minutes using all tools like Lego .. but none of the solutions make me really happy. Cheers Peter

1 0

Is every LUV member a LA member?
by Russell Coker 04 Feb '15

04 Feb '15

http://www.itwire.com/opinion-and-analysis/open-sauce/66827-linux-australia- member-numbers-down-by-30 Sam reports that LA membership is down. To be effective in lobbying LA needs to be able to claim a reasonable membership. Being a LA member costs nothing and only requires being on the announcement mailing list (which is low traffic). Please join LA everyone! -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/

9 11

Re: ISCSI export of local storage on virtualization servers
by Peter Ross 04 Feb '15

04 Feb '15

Thanks Avi, From: "Avi Miller" <avi.miller(a)gmail.com> >> On 4 Feb 2015, at 11:42 am, Peter Ross <Petros.Listig(a)fdrive.com.au> >> wrote: >> >> I have after installation a lot of unused disk space I would like to >> export for "other use" via iSCSI target. >> >> It is all looking as "borrowed" from RHEL 5 - so.. may an iSCSI target >> from there work? > > Technically: yes. However, we strongly recommend you don't. However, if > you absolutely really want to, I recommend using NFS instead of iSCSI to > share the unused space. Curious question: Why do you recommend this instead of iSCSI? > You should also be using Oracle VM 3.3 (which is the latest) which has an > OL6 userspace, not OL5. Yes, OL6. Sorry, got that wrong. BTW: Searching for possible iSCSI targets, I found this exotic one: http://rpmfind.net//linux/RPM/epel/6/i386/netbsd-iscsi-20080207-3.el6.i686.… > Also, with OVM 3.3, the unused local space can now > be used as a storage repository for local VMs. Yes. But it is on ocfs2.. and local. So no fail-over, and backup/mirroring for cold stand-by is all very clunky, it seems. > Again, ensure you're running Oracle VM 3.3 which uses MySQL Enterprise by > default. It's also a lot faster. Yes, I have this, and noticed that it did not like my local MySQL install. > However, the initial startup of WebLogic > can be pretty long as it runs all the discovery scripts to find and > enumerate active OVM servers. I think I should only start it when needed, as long as I have it on the local desktop. >> Location-wise, what is the "right thing to do"? I am tempted to put it >> on >> a Oracle VM Server but that would not work well if I have a Server >> problem >> and no manager either... Well, I could put it on two servers?? > > Either run it in VirtualBox or install it on another server. We don't > recommend running Oracle VM Manager as a guest of the server its managing > for exactly that reason. Yes.. At the moment my server setup is all a bit patch-work.. I have to "redesign" to make it nice and easy again. The exercise I am going through, satisfying proprietary software, is exactly a demonstration to show the benefits of open software because the later gives you the freedom of choice:-) Cheers Peter

2 1

ISCSI export of local storage on virtualization servers
by Peter Ross 04 Feb '15

04 Feb '15

Hi, I am "rediscovering" VMWare ESXi/vSphere (after 4 years things have changed a bit) and discovering Oracle VM at the moment. I have a question related to local storage on both.. I have after installation a lot of unused disk space I would like to export for "other use" via iSCSI target. For Oracle VM, there is a public yum repository but it contains iSCSI initiator stuff only. It is all looking as "borrowed" from RHEL 5 - so.. may an iSCSI target from there work? For VMWare ESXi, I did not come to that yet, I am just starting to investigate. If you have ideas, please let me know. As a side: I installed the Oracle VM Manager on my 8GB desktop, which made the boot so tiresome that in the beginning I thought, it is not booting at all (making coffee is too fast, it gives you time to drink it too) Location-wise, what is the "right thing to do"? I am tempted to put it on a Oracle VM Server but that would not work well if I have a Server problem and no manager either... Well, I could put it on two servers?? Regards Peter

2 1

QR code generator
by Andrew Greig 03 Feb '15

03 Feb '15

Hi all, Could I please get some advice concerning my failure to get the QtQR program running under SuSe 13.1 I have downloaded the rpm and installed qtqr as well as python-qrtools I have no joy from the GUI "run" process and from the cli I get andrewg@andrewg:~> /usr/bin/qtqr Traceback (most recent call last): File "/usr/bin/qtqr", line 15, in <module> from qrtools import QR ImportError: No module named qrtools andrewg@andrewg:~> I tried modprobe but that command no longer works. I have a project where I need to generate around 350 different QR codes, all of the same format. And then I need to print them in such a way as to recognise each label easily to correctly locate them. Thanks Andrew Greig

3 5

CoreOS and Ubuntu Snappy
by Toby Corkindale 03 Feb '15

03 Feb '15

For a Linux mailing list, it does seem very Debian-focussed here, with a little Fedora and Ubuntu. No-one has even mentioned CoreOS <https://coreos.com> or Ubuntu Snappy <https://developer.ubuntu.com/en/snappy/> here at all. If you haven't heard of those, it's worth having a look. They're both more focussed on server deployments, but they're switching around how packages and the OS are maintained. In both systems, the OS is stripped down, and comes as a single package, but is simple to update. (And in Snappy's case, also easy to rollback) Apps and libraries are rolled up into bigger chunks as well, removing dependencies mostly. CoreOS uses Docker images for everything; Snappy seems to use some of their own combination of cgroups+lxc+apparmor for separation, with mostly-statically-compiled apps. (Snappy says that developers can choose to statically link libraries, or they can choose to use system-installed versions of some "frameworks") Both these systems are interesting in the way they've decided to remove fine-grained packaging and updates. Having seen issues arise from version conflicts before, or just simply been fed up with having dozens of small updates a day, I can understand some of this motivation. I'm sure we've all been there before, when one package really, really wants a new version of an underlying library, but updating that library then requires updating other packages that maybe we'd have rather left alone.. Or consider that it solves another problem - you just want to try and have identical servers, but with conventional systems, by the time you finish running apt-get updates on a hundred of them sequentially, some of the packages will have been upgraded between when you started and finished.* With CoreOS or Snappy, it's much easier to just say "I have version 123 of the OS installed." You can test some servers with version +1 and then when they are good, you can push that specific version out to everything else. (Note: Disk space is cheap now, and the modern kernels deduplicate memory pages; so statically linking stuff in doesn't hurt storage or memory.) Anyway.. thought I'd mention these here since no-one else has been :) -Toby * Yes, using a deb cache or mirror alleviates this problem, until your servers are in separate data centres with independent caches or mirrors...

1 0