Choosing a VPS hosting
And so, since the rate of problems and issues of all kinds has slowed down and I could now breathe with relief, I’d like to share my story of picking out a VPS hosting. Please, don’t get me wrong and don’t treat this post as another commercial – it’s not.
As soon as the idea of having my personal blog had firmly stuck in my head, I started to look around for any means of achieving that. At first I tried to use blogger.com service but soon enough I realized that I wanted to have more control over my small spot of presence and to be responsible for everything that relates to running a personal web site. I wasn’t after anything special, nor I needed a help to configure the environment and, what was more important I didn’t intend to spend more than 20$ per month. That brought me to the single option – VPS or Virtual Private Server. Googling didn’t help much since the range of offerings and price tags was, and it’s still true, stunning. And all that just made me puzzled for a while. I spent sometime rambling through www.webhostingtalk.com forums looking for a hint. During one of such wanderings I came across a post that mentioned PhotonVPS, just a few words asking for a hands-on experience (I’d visited their site before and found them alluring but was afraid to burn my fingers) and since there wasn’t any negative reply I decided to try it myself. To tell the truth, the choice I made was 100% heart based but after one month I’m assured that it was absolutely a right choice.
See it yourself, I pay 11$ for 20GB of disk space, 500GB of traffic and 512MB dedicated (burstable to 1.5GB) RAM for a semi-managed VPS hosting with pledged 99.9% availability. The server was up and running in thirty minutes after I effected the payment and lately the response from the support team was supersonic and so far I haven’t had any outages. I only had a single issue going from Fedora 5 up to Fedora 10 when during the upgrade from Fedora 8 to Fedora 9 rpm database went corrupted to the point of no return. But even in that case the support team worked brilliantly and reinstalled everything from the scratch in less than an hour, noteworthy to mention, that before proceeding I was asked for the confirmation to double check that I didn’t mind to lose everything. Since by that time I hadn’t had anything valuable it was an easy decision.
In the end of the day, I’m truly satisfied and just want to express undying gratitude!
P.S. I hope this post won’t put a whammy on PhotonVPS ;-)
In: Life · Tagged with: hosting, vps
ZFS source tour
Whilst reading another highly interesting and cognitive post about ZFS SLOG from Brendan Gregg I came across a valuable link that holds a brief overview of ZFS related source code.
Devote a little part of your time to familiarize with ZFS structure and its main building blocks. It’s definitely worth that and, who knows, may be one day the time you spent will be repaid with interest.
Things to keep in mind when replacing a PS on SF6800/6900
What could be easier than swapping a power supply?! True, very true but not when you’re dealing with SF6800/6900. The problem rises from the fact that it’s quite tough to insert a PSU into its bay because of a tight fitting. And there is a chance to make a pause during the insertion that could lead to a momentary drop in the 56V power line inside the server because not all pins have been connected to the power centerplane. And as a result, say goodbye to all of your domains and be prepared to power cycle.
To avoid that an engineer should try to insert a power supply in one smooth and uninterruptible motion. Thankfully, this issue has ben resolved with power supply 300-1595-03 and with the release of power supply 300-1930-01. So if you’re dealing with one of them then you should be safe.
Refer to the following link for more details.
Oracle’s option that neglects DR on SUN
I’m in no way an Oracle expert and my opinion has no influence but still I’d like to point out on a ridiculously stupid behavior caused by a hidden Oracle’s parameter called _enable_NUMA_optimization. As I’ve been told by our DBAs, in Oracle 9 this option was set to FALSE but in Oracle 10.2.0.1 they decided to change its default value to TRUE. Not a big deal. A metalink note pledges that setting this option to TRUE improves the performance by splitting SGA into multiple segments (ipcs -a confirms that) and as a result reduces the total number of remote cache misses on a large system. But in the real life it’s not as glaring as it sounds. Conversely, with this parameter set to TRUE don’t even think about adding/removing SBs/CMUs into your big and beefy Sun box.
Per Oracle development, adding/removing processor groups are not supported. Oracle does not support dynamic reconfiguration with instance up while running on Solaris 5.9 or 10 when NUMA is enabled.
Isn’t that sweat?!
P.S. If you have a Metalink account then check Doc ID: 761156.1 for more details.
P.P.S. Update. Oracle has published a new alert, Doc ID 759565.1, that discuss their current guidance on NUMA optimization. They now discourage setting hidden Oracle parameters in favor of applying a patch that disables Oracle NUMA optimizations by default on 10.2.04 and 11.1.0.7. This update was taken from here.
Disable mpxio per vendor basis
If you ever wondered about disabling mpxio not only per port but per vendor basis then you’d be surprised to know how easy it’s really to fulfill. Just edit /kernel/drv/scsi_vhci.conf file appropriately:
device-type-scsi-options-list = "VendorID1ProductID1", "disable-option", "VendorID2ProductID2", "disable-option", ... "VendorIDnProductIDn", "disable-option"; disable-option = 0x7000000;
So if you, just like me, want to disable mpxio for Hitachi 9570 just add the following few lines to /kernel/drv/scsi_vhci.conf:
device-type-scsi-options-list = "HITACHI DF600F", "disable-option"; disable-option = 0x7000000;
Note. VendorID must be exactly 8 characters long, so if it’s not (Hiatchi is 7 characters in length) just add extra spaces.
vxdmpadm path activate weirdness
Just stumbled upon a strange behavior of vxdmpadm which requires further investigation.
The problem I’ve faced with during an attempt to set certain path “active” to loadbalance the data flow on HDS between its controllers. The built-in help clearly states that:
# vxdmpadm setattr help
vxdmpadm setattr path
pathtype=
pathtype can be either:
active
nomanual
nopreferred
preferred [priority=]
primary
secondary
standby
So I dully expected it to work as declared but instead I got the following error:
# vxdmpadm getsubpaths ctlr=c2 NAME STATE[A] PATH-TYPE[M] DMPNODENAME ENCLR-TYPE ENCLR-NAME ATTRS ================================================================================ c2t50060E800042A5F0d81s2 ENABLED(A) PRIMARY hds9500-alua0_0051 HDS9500-ALUA hds9500-alua0 - c2t50060E800042A5F3d81s2 ENABLED SECONDARY hds9500-alua0_0051 HDS9500-ALUA hds9500-alua0 - c2t50060E800042A5F0d82s2 ENABLED(A) SECONDARY hds9500-alua0_0052 HDS9500-ALUA hds9500-alua0 - c2t50060E800042A5F3d82s2 ENABLED PRIMARY hds9500-alua0_0052 HDS9500-ALUA hds9500-alua0 - # vxdmpadm getsubpaths ctlr=c3 NAME STATE[A] PATH-TYPE[M] DMPNODENAME ENCLR-TYPE ENCLR-NAME ATTRS ================================================================================ c3t50060E800042A5F1d81s2 ENABLED(A) PRIMARY hds9500-alua0_0051 HDS9500-ALUA hds9500-alua0 - c3t50060E800042A5F2d81s2 ENABLED SECONDARY hds9500-alua0_0051 HDS9500-ALUA hds9500-alua0 - c3t50060E800042A5F1d82s2 ENABLED(A) SECONDARY hds9500-alua0_0052 HDS9500-ALUA hds9500-alua0 - c3t50060E800042A5F2d82s2 ENABLED PRIMARY hds9500-alua0_0052 HDS9500-ALUA hds9500-alua0 - # vxdmpadm setattr path c2t50060E800042A5F3d82s2 pathtype=active VxVM vxdmpadm ERROR V-5-1-10357 Invalid argument or attribute specified.
Looks like I’ll need to investigate deeper to find the culprit but as a workaround just disabled the second path to force a failover to another one I tried to make active.
Update.
As always RTFM rules and I must admin that my apprehension, that with option one could change the state listed in the second column to active, was completely wrong. In the man page it’s lucidly written that pathtype=active is used to change a standby path to active.
# vxdmpadm setattr path c2t50060E800042A5F3d81s2 pathtype=standby # vxdmpadm getsubpaths NAME STATE[A] PATH-TYPE[M] DMPNODENAME ENCLR-NAME CTLR ATTRS ================================================================================ c2t50060E800042A5F3d81s2 ENABLED SECONDARY hds9500-alua0_0051 hds9500-alua0 c2 STANDBY # vxdmpadm setattr path c2t50060E800042A5F3d81s2 pathtype=active # vxdmpadm getsubpaths NAME STATE[A] PATH-TYPE[M] DMPNODENAME ENCLR-NAME CTLR ATTRS ================================================================================ c2t50060E800042A5F3d81s2 ENABLED SECONDARY hds9500-alua0_0051 hds9500-alua0 c2 -
Since in my case the path was already active it would be strange to make it active for the second time and as a result I got the error. So actually there were two options:
- Use vxdmpadm disable
- Use vxdmpadm setattr path pathtype=standby/active
So, folks, never underestimate the documentation. ;-)
In: Veritas · Tagged with: VxVM
Moving to my own domain
Just a quick update on what’s happening here ;-)
I’ve finally overcome my laziness and purchased a VPS, this is a subject for another post, quickly configured LAMP environment and installed WordPress so during the next few days or maybe even weeks, because currently I’m short on a free time, this site will be in constant appearance-related changing and polishing.
Have a nice day.
Rock is dead?
I do hope that at least this rumor is a pure speculation even it sounds convincing. Otherwise the future of Sun’s hardware business after the deal with Oracle is over looks increasingly dismal.
Data restoration from tape
Recently I had to restore some data from a tape written by means of Netbackup, so solely for the reference purposes I decided to write this short post.
First we need to mount the tape, I did this using robtest utility, and perfrom a robot’s inventory to make Netbackup aware about a new tape. Keep in mind that sometimes barcode visible on the tape itself could not much with what has been written on the tape during the backup. This discrepancy could be a result of different Netbackup’s barcode rules. To double check, use more e.g. more /dev/rmt/13cbn
After that I ran the following set of commands:
bpimport -create_db_info -id F006L1 -L /tmp/bpimport.log bplist -C client's_name -l -t 4 -R / bprestore -B -S source -D destination -C client's_name -t 0 \\ -L /tmp/bprestore.log -R /tmp/rename_file /what/to/restore
In: Veritas · Tagged with: NetBackup
Handful of links
Just came across a very useful blog note full of the essential Solaris papers. Solaris white album: essential papers. Noticed that some of these documents were lately released imprinted verbatim in “Solaris Internals” book.