Unix Admin Corner

Is Google server nodes the new Mainframe

noreply@blogger.com (jamesd_wi) — Thu, 12 Mar 2015 00:11:00 +0000

A friend mentioned that small/middle companies are using mainframes, but I think an argument could be made that Google and Facebook and friends are really creating the new mainframe with there various server nodes that are often made up of custom tasked designed hardware/servers.

One of the explanations of what made a mainframe a mainframe. I learned over the years was that a mainframe had custom hardware for specific components of the system, such as a hard disk controller that could be given a list of blocks or files and it would go fetch the data and place them into the system memory or returned to the program. Or a network controller that would do the same with network IO.

Has Google, Facebook and friends created a network by using one or more of their server nodes to make up what could be effectively be called a mainframe. For example if you consider 1 or 10, or even a 1000 server nodes running memcached a storage controller, the programmer can in a single function task the memcached servers with fetching and returning literally 1000's of requests for data all returned over the network with what many people consider the equivalent of drinking from a fire hose, because a 1000 nodes are essentially trying to return the data at wire speed. Various other technologies exist that allow other servers to return or store various chunks of data to other servers. Such as a restful API.

Since a mainframe is not limited to a single box or a size of box why not consider one rack of server nodes or even 1 or more rows of a data center a single system. Projects like Mesos and kebernetes allow the programmer to consider the full cluster of nodes a single system. Surely back in the 60's, 70's and 80's the mainframe were made from a lot of custom parts and not commodity parts, but by doing custom nodes, some of custom nodes are tuned for disk storage, others are tuned specifically for networking or ram based nodes, in the future they will be moving to GPU capable nodes, that have one or more GPU processors on board. Yes it would be the equivalent of making mainframe out of lego blocks.

First new post after a long rest...

noreply@blogger.com (jamesd_wi) — Thu, 12 Mar 2015 00:10:00 +0000

Well time to wake up this blog, will be posting stuff again, beware.

N40L awesome deal, my new toy

noreply@blogger.com (jamesd_wi) — Wed, 08 Feb 2012 04:28:00 +0000

Update:
its quiet enough to live in my living room, but it is currently in my basement, when I installed Nexenta it was sitting on my dining room table about 2 ft from my ears, and it was barely noticeable. It has a single low rpm 120mm fan,unless you somehow get a bad fan in yours, it should be fine.

http://www.newegg.com/Product/Product.aspx?Item=N82E16859107052

Cache Memory: 2GB
Memory Type: DDR3
MAX Memory Capacity: 8GB
Memory Features: 2 DIMM slots Unbuffered ECC
Model #: 658553-001
Item #:

$249^.99

The system last night, from NewEgg.com, along with the 8GB of ECC memory an extra $60 but still less than best price of the server. When I ordred mine the deal was for $299 and included a free 2TB hard drive, not quite good now, but the server is still an excellent choice, Nexenta Community version booted right up on it, had to install via a usb cdrom or dvdrom. Other wise there is a sata port for the for Optical drives and a power plug but the process of fishing the cable through the tight case and since I was only going to use it for installation I grabbed USB dvd-rom and gave it a try and it worked, smooth as silk the entire process. I will be working on adding a a second 2TB drive to the one that came with the system, and post benchmark numbers soon. I will try and do a better job keeping my blog up to date.

Node on Nexenta

noreply@blogger.com (jamesd_wi) — Wed, 17 Aug 2011 03:31:00 +0000

I discovered node.js that Joyent is using extensively to work its cloud magic. What is node.js check out this video for info, it seems very cool.

http://www.youtube.com/watch?v=jo_B4LTHi3I&feature=player_embedded#at=2825

Now the next question becomes how do I get it running on Nexenta. First hope was that it was already there considering the following announcement of the of the Joyent/Nexenta partnership

https://www.nexenta.com/corp/newsflashes/102-2011/918-joyent-and-nexenta-partnership

jamesd@amd:~/nodejs/node$ apt-cache search node | grep ^node

jamesd@amd:~/nodejs/node$

nope no luck, okay that was out, well since I wanted to give it a try I popped over to my Ubuntu box and then built node using http://www.codediesel.com/linux/installing-node-js-on-ubuntu-10-04/ and it worked fine, and was painless. Except the git node repository has moved to github.com/joyent/node.git

opteron:nodejs$ uname -av

Linux opteron.themagic.com 2.6.32-33-generic-pae #71-Ubuntu SMP Wed Jul 20 18:46:41 UTC 2011 i686 GNU/Linux

opteron:nodejs$ node

> process.platform

'linux'

Well now for a leap of faith... back to Nexenta and give it a try.

To make a short story even shorter, it WORKS!!! using eveything just the same.

jamesd@amd:~/nodejs/node$ uname -av

SunOS amd 5.11 NexentaOS_134f i86pc i386 i86pc Solaris

jamesd@amd:~/nodejs/node$ node

> process.platform

'sunos'

Moving onto Nexenta Community edition

noreply@blogger.com (jamesd_wi) — Thu, 05 May 2011 04:07:00 +0000

A while ago I manged to screw up my ZFS pool by having it use files on other ZFS pools as cache and slog devices. After 6 different versions/distrobution/releases, I have given up, and recreated pools that are much saner in layout.

The filesystem now runs Nexenta community edition

With the following pools
250GB root pool or syspool as Nexenta calls it
3x 500GB in a mirrored layout in a pool called tank I may break it into a 2 way mirror later. And then create another 2x 500GB pool or zdev on the pool, currently this pool is mostly being used for ESX nfs storage. I wanted to go 3 drive wide mirror for read speed and improved write performance over raidz.

I was going to use SXCE on the system but during the install it hung at 18%, so I gave up and went with Nexenta since I had the disc sitting next to the machine anyway. I am enjoying it mostly did have to add another repository to get bind9 build for it, but that plus about 15 minutes of configuring it I was able to get nexenta to be a DNS server, thus making NFS happy.

Future:

Well I don’t expect to be using this fileserver long, since I want to move to a all in one box solution for ESXi+ZFS using an HP ML350g6, but this will allow me to add more storage to my existing ESXi 3.5 box, its built in 60GB is showing its age with its SCSI drives that are expensive for anything larger than 36GB and use lots of power in return. I will be picking up components to go in the ml350 that can live in this box till the main (expensive) bits are ready.
I will probably get 4x 2TB drives and put them in this box perhaps even the ssd’s for l2arc/slog can go in this box as well.

Success!!!

noreply@blogger.com (jamesd_wi) — Sun, 17 Apr 2011 16:53:00 +0000

Well after 4 days of struggles and unexpected patching sessions, I now have my new 750GB SATA drive in my laptop. Once I started using samba it all went much better than I expected. To make things faster be sure to choose restore partition not files. Restoring files takes a lot more time. Despite your first guess that restoring by partition doesn’t require extra steps if you installed a larger drive, it uses the partition scheme you setup on your new drive. You just need to specify where you want your files restored too. The process goes much faster, when I tried using the file method it gave me a time of 19 hours, and went up occasionally to 3-5 days. When I switched over to partitions in the morning when the timer still gave an ETA of 18 hours, it went down to 2 hours, but took like 3.5 hours to complete the process.

When I rebooted the laptop after the restore completed, it immediately started booting Windows, no hiccups, no little tweaks, it all just worked, I was surprised and impressed at how well it did the job. Of course this raises the question could I have used full drive backup and got better performance, and still restored to a larger partition?

After windows booted, it did have to look for and install a driver for the hard disk, I didn’t change the controller, so I was surprised it needed to install a driver for a SATA hard drive that replaced a SATA hard disk? But oh well it worked, I will have to see next time I reboot if it gives any performance enhancement.

The reason I purchased Acronis and didn’t use Clonezilla and gparted type solution was because I was afraid that I wouldn’t be able to make Vista boot on the new disk, all the entire how-tos I read required repair work and/or recover disks. My laptop didn’t come with a Vista install disc; I did create the recovery disks nearly 3 years ago, damn if I know where those disks are hiding. So in the end Acronis did what I expected just the backup progress wasn’t as easy as I expected, but the restore pretty much exceeded my expectations by far.

But it is nice to see 487GB of free space, and even gave 60GB of space to Linux partition should I choose to dual boot sometime in the future.

Lessons learned in this process.

Install as Administrator, an administrator enabled account isn’t enough, after the restore process it still doesn’t like running as a normal user.
Don’t bother using FTP with Acronis it really doesn’t play well with others.
Defragment your drive before you start, backups go faster with it defragged.
Use the fastest transport media possible, wired is much better than WIFI even “N” class wireless
Gigabit would of helped, but wasn’t an option on this laptop
Don’t use dynamic style disks the restore process doesn’t like it, and you can change later
Be sure to have another computer if possible for several days while the backup/restore is proceeding.
Install gkrellm on your fileserver so you can see throughput, and be sure that Acronis is doing something.

Progress

noreply@blogger.com (jamesd_wi) — Sun, 17 Apr 2011 03:30:00 +0000

I was never able to make Acronis Trueimage 2011 complete a full backup to any Linux based FTP server, I finally just found the simplest samba configuration document and created a samba share, which worked painlessly. I even took the extra couple hours for it to complete a full validation of the backup before beginning the drive replacement procedure. Not sure if I read something for an older version about not using samba for the backup or what, it was completely painless except for the fact that it took all the way to Sunday to complete a full backup of a 180GB disk. Something should have been done in 8 hours.

I bought the Acronis Disk Manager Home along with the Trueimage home, it was about $65 combined for the together not bad, it made the job of formating my new 750GB 2.5” SATA disk go smoothly and allowed me to setup a Linux partitions for ext3 and swap. BTW if you are in the market for laptop hard drives, the current 1TB drives are 12mm in height and most laptops can only take a 9mm tall drive so don't order the 1TB drive that you really think you need before you verify it will fit in your laptop and save your self the return and re-order hassle.

I sit here and type this as the recover operation runs so we shall see how this all turns out in a few hours. Its only reading data off the samba server at about 7-8MB/s compared to 12MB/s that it wrote at. I guess Linux based recover program is slower than the window hard disk based version. I'm pretty sure the new hard disk is faster than the old one it replaced and the 1.5TB disk in the Sun Ultra 20 with dual core 2.6ghz Opteron should be up to the task of filling a 100mbit ethernet to the laptop, the ulra 20 has a Gigabit Ethernet port and is plugged into my gigabit wifi router.

The process has been going on for about 30-40minutes, and it still hasn't provided any hints as to how long the recover process is going to take, thankful I have gkrellm running on the Ultra 20 so I can see that it is sending between 1.5 and 6MB/s across the network. It still doesn't appear to have any color to the progress bar on the Acronis recovery dialog box.

I guess I'm going to take my own advice in these situations, walk away and let the computer do what it does best.

UPDATE: well after about an hour, it did give me an ETA for completion, I really don't like the number, 19, 19 hours but what can I do. But its the story of this progress everything takes much longer than I think it should!

This Close!!!

noreply@blogger.com (jamesd_wi) — Sat, 16 Apr 2011 05:02:00 +0000

I decided to buy TrueImage 2011 home, to backup my laptop, since I will be replacing its hard disk, I am seriously thinking of returning the software and asking for a refund, it’s been a total pain, first issue had to do with permissions and it not liking the software I had installed previously not sure which software, but I had to disable my virus checking software, and every service not written by Microsoft, and then run it as “Administrator” not sure which one of those things killed the initial startup, but I did get it going finally, But that is only the first stumbling block.

I am using the ftp destination for backing up, which basically goes like, first time I tried it with just plain wireless, N class network adapter and after it uploaded its first gigabyte it estimates the ETA of 2 days and 8 hours. That would never do. So I hook up wire and see if it could recover, it can’t FAIL. Like really it can’t restart a ftp connection? How lame.

But it gets better. I restart the backup process, this time it gives a better speed only 8 hours to backup the 160GB hard drive. Well it’s now about 1am, and I have to work in the morning, so I lock my screen and head off to bed. I wake up in the morning thinking it will be done or perhaps I will have an hour or two left. You guessed it, it’s not done, I sit looking at my login screen, Microsoft decided it was the right night to install patches, and reboot my machine around 2am…. Not sure how much I blame Acronis for this issue, but it could of made Microsoft wait… I kick the backup off again and head off to work. I get home, to find an error message and the backup has failed, used their trouble shooting tools that link into their knowledge base, a few hits, but nothing seems valid, so I try again, and 8 hours more, FAIL again. Lots of fighting with ftp connections, sure you can create them, but they don’t give you a way to delete them and start over, talking about second class destination.

Well this process started Wednesday at around 6pm, it is now Friday 11:39pm, I finally got tired of fighting this and decided to open a ticket, I avoided this process because it feels too much like being at work opening tickets, of course there automated system decides to use Microsoft outlook to send off an email, I have outlook installed but not configured so fail again. I think spent another 20 minutes looking for the way to upload it manually. After I upload there “troubleshooting packet” the upload process tells me it knows a possible solution, so I read the first link and low and behold comes a page that mentions it doesn’t like proftpd, that I have been trying to use, okay you know your software hates proftpd and needs features that some ftp servers don’t support why not check for them and bitch about the lack of them preferably in pure English FAIL again. Further the doc doesn’t mention a ftp server that they support FAIL again. If you Google Acronis ftp server, you get there software to backup Linux servers. So we have a company that supports Linux, but won’t give you a list of ftpd servers that works in Linux with Acronis trueimage and I couldn’t even find a list of supported ftp servers on their forum. It seems like there whole error reporting is for machines, every error is a hex code, and can’t you print out the error means in English. This is a home product…

For those that found this blog entry because you have been bitten by the ftp problem. I did find an ftp server that appears to work, it is pure-ftpd which ubuntu has, all you have to do is install it.

All in One?

noreply@blogger.com (jamesd_wi) — Sun, 10 Apr 2011 01:19:00 +0000

Well the debate continues in an earlier blog post I discussed my thoughts on my new servers for fileserver and esxi box (debating new home hardware) my current thought has now changed from getting two servers into just getting one but running Solaris ZFS in a guest on it. I know ZFS loves memory, so to make Solaris or Nexenta ZFS perform its best I would get one HP ml350 g6 but upgrade it into the middle of its capacity, 2x quad core e5606 (3.13 ghz 4MB of l3 cache), and 4GB of memory from hp, (2x 2GB dimms from hp) and then 4x 8GB dimms from a 3^rd party memory supplier for a total of 36GB memory. For the controller I will use the p410/nz and grab a 512MB cache with battery backup to get full write performance for the disks I will add 2x 250GB from HP to hold Solaris and a couple other guests, and 4x 2TB sata drives from a 3^rd party in a raid 1+0 layout, two 460 watt psus. Even with the extra 8GB of memory and extra power supply it works out to be $800 less and more importantly it also uses less power so lower electric bill long term.

I think the above configuration should be a good enough performance. But if it still needs more performance Napp-it all in one, documents using a SAS controller in pci pass through mode with good results. The card they recommend is LSI 1086E SAS controller that Amazon has on sale for $160 but then I may need to get a separate disk box or put the disks in the non hot-swap bays. I don’t think my storage needs require overly fast performance and I could easily dedicate 8-12GB of memory to the Solaris/Nexenta guest perhaps even more. Third party 8GB dimms are only $270 and the ml350 g6 has 18 dimm slots so its not like I’m going to run out or I could possibly get a SSD device and attach to the p410 controller and pass that into Solaris/Nexenta guest.

Loving Autohotkey

noreply@blogger.com (jamesd_wi) — Sat, 09 Apr 2011 14:34:00 +0000

As a Unix administrator, I try and use my keyboard more than my mouse, and I love automating things and processes, until I found Autohotkey I really didn’t have a good way to do it in windows, sure you can record marcos or write small scripts in office basic or whatever they call it now, but that is limiting, or I could write scripts in virtualbasic , too much work.
Autohotkey really makes automating things like passwd entry, I know it’s a “bad thing” but when you are a new job site, and they really don’t understand the easy, power, security of ssh shared keys but they accept autohotkey what can you do? When in rome I say.

It would be nice if Autohotkey had a tool that allowed you to encrypt a string decrypt it easily so that you wouldn’t have to put passwords unencrypted in your script files, sure someone that really wanted access could crack them but it would at least keep the casual person from reading all my passwords.

Something like
myPasswd=”#$@SSGFSRWGSsses”
Send unencrypt(“%myPasswd”)

Here is my current autokey script on my home system.

#s::
InputBox, SearchTerm, Search
if SearchTerm <> ""
Run chrome.exe http://google.com/search?q=%SearchTerm%&ie=utf-8
return

#w::run winword.exe
#e::run excel.exe
#m::run chrome.exe https://mail.google.com/mail

Music Cloud

noreply@blogger.com (jamesd_wi) — Wed, 06 Apr 2011 12:25:00 +0000

If you haven't heard amazon won the race to the cloud for music, you start with 5GB for free download any album and get up graded to 20GB, and they dont count amazon purchases towards your quota. Now to get you started they have a a few good deals including four 99 mp3 albums for 99cents. I grabbed those four albums and motley crue's greatest hits for $3.99, now I can have music at work without having to remember to bring my mp3 player everyday. It so far works in google chrome, and will try internet explorer and firefox at work today.

Seed your cloud

Audio Geeks should check this out.

noreply@blogger.com (jamesd_wi) — Wed, 16 Mar 2011 14:30:00 +0000

audio geeks out there should check out http://www.vinylphilemag.com/pdf/vinylphile-005.pdf its a free publication done by a friend of mine, Rich Teer, it comes out monthly even if you don't care for vinyl its a worthwhile read for the audio gear stories.

Debating new home hardware

noreply@blogger.com (jamesd_wi) — Tue, 15 Mar 2011 18:55:00 +0000

Thinking about buying a couple new servers for home, a new ESXi 4.1 server and aSolaris ZFS fileserver, Here is what I decided the requirements of the first server are.

ESXi server requirements
Fully supported on the VMware HCL
Dual socket, quad cores 4MB l2 caches per proc
8+GB of ram, system needs to support at least 32GB for future expandsionHardware raid controller (makes esxi happier)
At least 4x 3.5” sata drives (may consider 2.5” drives)
Dual gigabit Ethernet port.
3 years next day business hours hardware support
Out of band management nice to have

I have been doing computers for almost 30 years in one form or another building x86 gear since 1985, and frankly I’m tired of working on my home hardware, I want something that is pretty much plugNplay I don’t want to mess with the hardware and I don’t want to have to worry about replacing the system for 3 years by that point there should be some more interesting things going on. I am leaning towards a standard tower server they tend to be quieter than there rack mount cousins and don’t feel the need to pay the rack mount tax as well. I currently have a dl380 g3 that is very loud and has no slow fan speed mode at least in ESXi 3.5. The current sweet spot for cpus are quad core 2.13ghz, slower only saves $20 and faster is a price jump of over $100 per cpu. Upgrading my current ESX server will also allow me to run 64bit OSes and will give me the excuse to power down another server as well for the most part.

I looked at Supermicro systems but I don’t see much price difference between them and HP that offers everything from the system to full support perhaps I am not finding the right systems. Supermicro offers some systems with lots and lots of disks, which I don’t need for home. Also lacks support and enterprise out of band management features of HP.

I also looked at Dell but perhaps I didn’t find the comparable model but I really didn’t find a lower price for my configuration, and Dell seems to be a slightly less of a reputation with less enterprise quality add-ons so less stuff to play with. So Dell is out of the running.

I looked at Sun/Oracle x86 options but they seem to be limited to rackmount gear that only supports 2.5” sas drives and the prices were way more than I am looking at and there online configuration tool is offline so I couldn’t reduce the ram and drive size to see if that helps the price. I really want to go with a tower based system because they are quieter. Cheapest x86 server I can find is over $3000, almost double what I have budgeted.

The purely beige box solution seems appealing because of price but it is hard to find a website that offers complete configuration of systems with dual sockets and they will also lack high quality support and the enterprise out of band management features. I really don’t want to assemble my own hardware.
So far my choices for the ESXi server are either HP Proliant ml150 g6 or ml350 g6 I am leaning towards the ml350 because it comes out of the box with a 3year warranty and it doesn’t require extra licenses to support the same configurations that I may upgrade to in the future it’s a bit more but you can expand it further. With the ml150 I need to get the hot plug model so I get a raid controller that works with ESXi out of box, the low end integrated controller isn’t supported by ESXi.

The shocking things with my configuration at the HP website is that the ml350 is cheaper than the ml150, among other differences the ml350g5 supports both the e5606 and e5506 cpus as opposed to just the e5506 that the ml150 g6 supports, the e5606 is cheaper. The specs are the same for these two cpus, same speed, same l2 cache, same watts, after looking at the Intel site I see the e5506 uses 32nm press as opposed to the 45nm that the e5606 uses. Which I see on various websites not to add anything except slightly lower power usage, the e5606 is fine for me. The low power cpus are too expensive I don’t think I would save enough to make them a reasonable choice. I keep debating on whether to wait on the second cpu, I like the idea to buy it day one so that it comes installed and fully tested, I’m sure it will add some speed and memory expandability but I won’t be using all 8 cores for a while. I never really found dual 3.2 GHz p4 in my old esxi server to be limiting it was always about memory.

Adding the redundant hot swap fans seem to be a good choice, an extra $50 for security having a fan fail while at work or while away on an extended vacation which is the time most hardware seems to fail. Would lead to a shortened life span of the server perhaps keeping it cooler all along I’m not going to keep the server in an air conditioned space it will be in my basement that may see temperatures in the 80’s perhaps even 90’s a few days a year.

Both servers support more than enough drives I will be purchasing one drive from HP for ESXi probably a 250GB sata drive $90, the extra $20 compared the 160GB drives is worth it, Larger drives from HP are too expensive so I will be adding standard home grade drives, and buying enough empty drive spuds on ebay to add 3 more disks, a second 250GB for mirroring of the primary drive and then 2x 2TB SATA drives for storage of guests all configured in mirrored layout for data protection and my low usage level should make the higher enterprise drives unnecessary. These later drives may be added later. Another downfall of the ml150g6 is that if I add the second drive cage increasing the internal compacity to 8 drives from 4, I would need to upgrade to the 750watt psu a $175 upgrade plus the enablement kit a $200 increase which seems really excessive and pretty much puts the final nail in the ml150g6 configuration in this comparison. I don’t see a need for all 8x drives but I do like the option of doing it later. I have thought about going with sas drives but they are a bit more expensive for 3^rd party drives and way more expensive from HP and are limited to 750GB currently. Frankly I don’t think I will see much speed difference because I won’t be pushing them hard with my work load, how much performance do 5 part time users need, and I can add more cache memory to the controller as well.

I would most likely be ordering the server with 2x 2GB dims for memory one dimm on each cpu as required by the hp website again the extra $20 between the 1 and 2GB dimms is worth it, and will be adding more 4GB dims from a 3^rd party memory supplier, they are about $100 each and ½ the price of HP’s memory and I can remove them should it become an issue for service personel.

I will only be purchasing one 460watt power supply and possibly adding the second power supply off eBay later for redundancy.

Dual gigabit Ethernet is important because I can allocate one port to iscsi/nfs IO and the other used for regular network traffic . The ml350 has dual gigabit standard, the ml150 would need a pci-e card. Long term I may even decide to add a 3^rd gigabit port the cards are cheap and getting cheaper by the day.

I also want hardware support for 3 years, by then I will be ready to replace the server, if I am not it should be at the point the where parts are dirt cheap and the parts should be easy to find no reason to pay more for a longer contract. The ml350 includes 3 years support, on the ml150 it’s a $188 dollar add-on. Adding the 4^th year increases the price $280 which just too much for a server that old.

So in the end the ml350 wins because its nearly the same price out of box and then all the extras add up to making the ml150 more expensive and less expandable in the end.

Server two, Solaris ZFS server

Dual socket (because expandability is nice) but only one cpu installed

6x 3.5” drives, 2x small drives for OS and 4x 3^rd party 2TB drives in raid 1+0 for a total usable of 4TB before compression

SSD for l2arc and slog 80GB added sometime in the future.

Dual gigabit or more
3 year support.

This server comes out a bit less since it only has one cpu, quad core because dual core options don’t save much money anyway.

I will buy with only 4GB on board memory and then add 3-4 more 4GB 3^rd party memory mostly used for ZFS caching.

250GB sata drive purchased with the system, with 1 more 250GB later and 4x 2TB drives added for home storage, less than $100 for 50% more security and increased performance I will also try to buy from 2 different drive vendors for a slight increase of potential security because less chances to up with a bad drive lot taking out the zpool.

All the other arguments remain the same.

Now the question I have is, how much of a discount can I get if I purchase from a HP reseller, my two servers I will not be enough to get me a big discount but a 5% discount would be nice. Anyone know of HP reseller that give discounts and preferably local to the Milwaukee, WI area.

why not to become a netadmin

noreply@blogger.com (jamesd_wi) — Fri, 11 Mar 2011 01:11:00 +0000

Every Network admin's nightmare 5days of no sleep and having to revert back to paper for all the major actions of the company.

Now this is one major torrent

noreply@blogger.com (jamesd_wi) — Fri, 29 Oct 2010 04:00:00 +0000

900GB of ancient web content, might be interesting to see, not sure I would want to give up that much space to do it, but it is only 1/2 a modern 2TB. I think some stuff I posted somewhere ended up being posted on a geocities site, so may be interested, maybe google will archive it and make money on it, and show yahoo how ads are profitable even if yahoo diddn't want to devote the servers to it.

Status update, gave up with the process.

noreply@blogger.com (jamesd_wi) — Tue, 26 Oct 2010 02:58:00 +0000

Well after over 3weeks of no progress, I just gave up, and powered down the system, I will give it another try when a new version of OpenIndiana is released. At that time I will setup for a long process, enable ssh, and possibly screen so that I can run multiple processes. Of course with my luck it will take 5 minutes ;-)

Perhaps lost in a corner

noreply@blogger.com (jamesd_wi) — Wed, 13 Oct 2010 05:26:00 +0000

Well I still let the import process continue, not sure if its proceeding, but the disk access led continues to blink, and blink I hope it will finish sometime.. I will have to set an end, but I haven’t yet.

To kill or not to kill

noreply@blogger.com (jamesd_wi) — Mon, 04 Oct 2010 15:16:00 +0000

About 2 weeks ago I attempted to import a pool (4x 500GB SATA drives) with zvol based slogs devices that were having issues in b134 after repeated hard crashes due to power issues. This attempt was done using a live Open Indiana DVD, after I imported the first two pools on the system I executed the import of the pool that I hoped would complete successfully this time. Even though It failed with the older versions.

Well not knowing or expecting that this import would take more than an hour or two, perhaps a day, I didn’t do anything to prepare the system to be in a state that would allow for monitoring theimport process. It was booted into text console as opposed to the default X gui based console, and sshd was not started. Thus I had no way to monitor the system that was stuck in the kernel, it appears to be progressing the disk activity was sporadic, barely noticeable only by the blink of the amber LED, some blinks were just a quick flash occasionally some were a flash that lasted a bit longer sometimes they occurred every couple seconds, some time they go 5 or 6 seconds between brief flashes but since the system was running on a live image with no writes or reads to hard disks to light up the disk access LED so it obviously working on the command I gave it, I had even pressed ctrl-z in hopes of pausing the command in a way that it could restarted in the background later and giving me the chance to run a command or two to monitor the system progress or at least what is happening going on.

But two weeks later the process continues only giving me two options, let it continue trying to work its magic and hopefully end successfully and giving me the chance to remote the zvol based slog devices from the pool or even just get a copy of the contents of the pool stored on valid pool or a different system with the required space. Or do I give up and shut down the box only to start it a way that would give me the chance to monitor the process and/or remove some zvol’s that aren’t par t of the pool hopefully speeding the process the next time in hopes of speeding the process.

I guess I will let it go for a few more days in hope that it will at least show some sign of completing shortly,

OpenIndiana Update

noreply@blogger.com (jamesd_wi) — Sat, 25 Sep 2010 13:42:00 +0000

I have ran up to ONNV b145 on this system, and was able to run “zpool import” scan without issue,but for some reason a single drive in the pool was causing it to crash with OpenIndiana b147. It didn't matter if the drive was on the on-board controller or the added realtek controller. Just the existence of the drive caused the crash, I didn't notice any scsi errors. And I had not recently.

Some background on why I want to move to OpenIndiana a few versions ago, I heard that you could remove slog devices, so I decided it was safe enough to use a zvol from another pool (one made up of a single 1TB drive) as a slog device making the raidz pool much more usable for nfs based vmware ESXi storage, what I didn't relive at the time was that yes you can remove a slog device from an imported pool, but should the slog device die or the underlying zvol get wedged somehow it wasn't possible to import the pool. Of course a weeks after adding the zvol based slog device the pool refused to import after a lightning storm came through one evening and caused power to fail 5 or 6 times in a couple hours.The zvol became wedged, can't snapshot or clone the zvol or use it there may be fixes for this in post b145 builds.

Now as of b146 it is possible to import a pool with missing slog devices. This should allow me to get my 1.5TB pool of mp3's, movies along with all the data files back on line. If you choose to follow my bad example and use a zvol as a slog device, I dont recommend it, the key to importing the pool is.

zpool import -m -f -d /dev/dsk/ -d /dev/zvol/dsk/[slog zvol pool]/   poolname    

Dont expect the above command to finish quicky, I started it around 10 hours, ago and it hasn't returned though I still see infrequent disk access on the system running openindiana live dvd. As soon as it returns hopefully successfully I will be backing up important data, and probably recreating the pool without the zvol slog devices.

OpenIndiana

noreply@blogger.com (jamesd_wi) — Fri, 24 Sep 2010 18:49:00 +0000

Tried OpenIndiana over the last few days, well I tried too anyway, first attempt was using there “pkg” enabled upgrade process, but that didn’t work because somewhere along the like I fubar’d something(s) over the years, even went back to my initial boot environment and gave that a try it failed as well.

Next came the USB drive method, the write process seemed to go well, but when booted I got a simple message on the screen “GRUB” nothing else no chance to edit or change anything. No boot happened the box was hung.

So decided to give the Live DVD a try, it burned and booted as expected, but then the real test came, importing my zpools, well I typed pfexec zpool import, then the disk light came on and 2 seconds later the box crashed and rebooted. I tried all the grub choices but still no luck.

This is a pretty run of the meal beige box, asus m2a-vm amd x5200+ dual core cpu and 6GB of ram, it ran b134 and b144 fine no issues, b147 doesn’t like it, I do have a realtek sata -> pci board in it that I will try removing later and see if that is causing the issue but that is a project for another day.

Checking Dolly’s Paternity

noreply@blogger.com (jamesd_wi) — Tue, 22 Jun 2010 14:50:00 +0000

Here is a one liner to print out which file systems and zvols in your ZFS pools are clones and print out its parent.


zfs list -s origin -r -o name,used,refer,origin  | sed -n '/-$/d;p'

Steps to Clone a VMWARE guest on a ZFS NFS share

noreply@blogger.com (jamesd_wi) — Thu, 27 May 2010 00:59:00 +0000

Today’s task was to create a base image to use for Windows on my ESXi box so I don’t have to sit through windows install everytime I want to play with windows apps. Of course being a starter ESXi system I don’t have vCenter or anything else that provides vMotion. Here are the steps.

Create a ZFS filesystem and share it with root read/write permissions.
Set your favorite zfs options, disable atime, enable compression, etc.
Setup nfs sharing with root privs

zfs set sharenfs=root=@192.168.1/24 tank/esx2 # please make as secure as you wish.
zfs set quota=20g tank/esx2
# unless your cpu is really slow I think compression would be great because all of these
#blocks will be shared among the clones
zfs set compression=on tank/esx2
zfs set atime=no tank/esx2

Go to your ESX(i) box and create a new dataset using the newly created share.

Create a new guest storing everything on the new datastore.

First I installed 2k3 enterprise installed 16GB harddisk ( yes way too big, but if you don’t use the space its basically free, and it’s a pain to run out).

Then I installed all my favorite applications and plugins so I only have to do it once.

Vmware tools
Filezilla
Firefox
Isomagic
Putty
Adobe reader
SNMP ( nice to add this so it’s easier to monitor)
Flash Player
Notepad++

Now configure a few things, so we do it once and don’t have to do it again.
Set hostname to be “baseimage” or something.
Make sure the guest is set to use dhcp so you don’t have conflicts later

Create user accounts
Configure proxy
Configure web homepage
Configure snmp
Configure ntp

the next steps I’m not sure how necessary they are but I figure I would do them anyway.

Clean up C: drive
DEFRAG DISK

Now poweroff the guest

Login to your OpenSolaris box, take a snapshot of the guest filesystem

zfs snapshot tank/esx2@base_windows_image
zfs clone tank/esx2@base_windows_image tank/new_guest
zfs set sharenfs=root=@192.168.1/24

rename the directory that holds your base image to something related to your next guest, changing the directory name won’t use more than 1 block of space unless you give it a really long name.

Now back to the esxi box and add another datastore with the cloned directory

Create a new guest using custom method, use an existing image
Choose the one on the freshly cloned filesystem
Add/remove hardware as you desire, does anyone use floppy disks anymore?

Boot it up everything should work

Now reset the hostname to what you need it to be.
Give it a static ip if desired.

Customize the guest as you see fit…

Done

The ZFS pipe is clogged

noreply@blogger.com (jamesd_wi) — Mon, 24 May 2010 00:56:00 +0000

Here are the changes that are stuck in the ON pipeline waiting for Stable OpenSolaris 2010 to get out the door. A lot of performance improvements and usablility stuff, I'm sure I missed a few.

b135
zio_ddt_free is single threaded with performance impact
zil need not inflate blocksize that much
slog can probably pack 2X more data per lwb
dsl_sync_task_group_wait() can wait too long

b136
old iscsi target dies...
BUG/RFE:6922272SATA framework does not handle greater that 2TiB disks

b137
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6917066

b138
For 4k sector support, ZFS needs to use DKIOCGMEDIAINFOEXT

b140
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6280630
add interval (count) args to zpool list
dbuf_read_done() locking performance improvement
RFE: provide interval arg to zpool status to monitor resilvering
COMSTAR should avoid extra data copy to zvol-based backing store

DeDup will get faster on lowend machines

noreply@blogger.com (jamesd_wi) — Sun, 23 May 2010 23:46:00 +0000

At the end of Dedup Status thread is a cool little tidbit that I missed but it gives all us Home OpenSolaris users that can't afford expensive SSD's something to wait for, It is putback in b141, and as that the wait for the next OpenSolaris stable has put all /dev releases on hold we all get to wait, hopefully.. they jump a few releases when stable opensolaris is released so we don't have to wait 3 months to get this fix. There are bunch of other fixes that went into OpenSolaris /dev that the non hard core guys that don't build there own releaes have to wait to get as well.

George Wilson wrote:

I just integrated a performance improvement for dedup which will
dramatically help when the dedup table does not fit in memory. For more
details take a look at:

6938089 dedup-induced latency causes FC initiator logouts/FC port resets

This will improve performance for such tasks as rm-ing files in a dedup
enabled dataset, and destroying a dedup enabled dataset. It's still a
best practice to size your system accordingly such that the dedup table
can stay resident in the ARC or L2ARC.

- George

Feeling Old yet?

noreply@blogger.com (jamesd_wi) — Sun, 23 May 2010 17:13:00 +0000

PAC-MAN turned 30 this week.... Spent hours playing this game as a teenager, had it on my Atari 2600 and played at the arcade...