logo
IRC Archive / Freenode / #solaris / 2010 / February / 25 / 1
loko
Should you be using overlap for x86 or not?
axisys
previous topic: compression on or off
new topic: hw raid cntrl or zfs raid ?
loko
hardware
axisys
i have a t5410 (i think)
i prefer zfs raid w/ ssd for cache and log devices ..
i think it will perform better if i have enough memory .. since hw raid control do not have as much mem
sorry t5140
MilkmanDan
If your hardware "raid control" has any memory in it at all it'd damned well better be battery backed, or you ought to disable it immediately.
I can't understand how proper Unix shops that have years of institutional knowledge in SVM or VxVM root mirroring can suddenly decide that "hardware RAID" for OS disks is a good idea.
peerce
yeah, the main performance gain of hardware raid is write performance if you have battery backed write-back cache
MilkmanDan
If you're only paying for VxVM for OS mirroring, yeah, ok, I can see a justification for saving the money.
But nobody in their right mind has done that this century.
axisys
peerce: so with ssh for log and cace wont outperform the hw raid write-cache ?
s/ssh/ssd/
s/cace/cache/
peerce
even SSD has write performance issues. a write cache means the host can treat hte data as committed as soon as its sent to the controller
axisys
peerce: but with cache zfs can do the writing asynchronous .. to should not introduce any delay either.. no ?
i am just trying to understand.. not arguing w/ anyone
peerce
axisys; cache in main memory is toast if there is a power event, or system crash
axisys
peerce: no i meant if i have ssd cache device ..
peerce
but that takes longer to write to tahn a raid controller BBU write-cache
Stric
(l2arc read) cache device, or zil/slog?
axisys
Stric: zil/slog
gothos
peerce: I fixed my DSEE7 problem, more or less, I set cert-pwd-prompt to off and can now start the DSEE7 just fine
axisys
peerce: is that right? btw, whats BBU ?
Stric
battery backup unit
axisys
Stric: thanks..
Stric
or backed
mini-UPS
axisys
peerce: hmm.. did not know it takes longer to write slog on ssd than on raid controller w/ bbu
Stric
dram is fast at writing, disk and flash isn't
flash units usually "fix" this by having some dram where your writes end up
so if you rely on that to be written, then you can be toast
axisys
Stric: oh yeah.. raid controller has real memory.. but if the total mem is slow it has to buffer somewhere else in slower location
but if the toal mem is full..
Stric
I don't follow
a raid controller with battery backed write cache memory, it puts writes into dram and then tells the host "ok".. and starts writing that to the slower disks in the background
axisys
if i have big chunk of data to write commit .. raid controller memory wont have enough of it to hold .. so it has to keep those cached in slower disk or ssd anyways.. no?
Stric
raid controller memory is usually 128MB, 512MB or similar
nothing is "cached in slower disk".. either it's cached in fast ram, or stored permanently on slower disk
axisys
Stric: ok.. so performance wise raid controller seems a better choice ..
Stric
could be
unless that raid controller sucks
and that set of controllers is not empty
axisys
Stric: well can you elaborate? i am going to be using t5140
Stric
separate raid controller or something built-in? if the built-in, then it doesn't have a battery backed write cache, so this whole disscussion is moot
axisys
Integrated hard drive controller supports RAID 0 and RAID 1.
http://sunsolve.sun.com/handbook_private/validateUser.do?target=Systems/SE_T5140/spec&source=
Stric
and no battery backed write cache, so no write cache
so all it does is raid1 with less protection than zfs
axisys
it has option to buy pci raid controller to expand
the site will not have power outage.. has generator
Stric
the OS can panic
someone can trip on the power cord
with proper stuff, neither of those matter
axisys
Stric: gotcha
Lithium Ion Battery Module, 8-Port SAS Adaptec RAID HBA, RoHS:Y , SGXPCIESAS-R-INT-Z
that is a pci card to buy as optional to expand
internal i think
this one shows zfs raid gave more iops that hardware raid http://storagemojo.com/2006/08/15/zfs-performance-versus-hardware-raid/
may be v440 raid controller sucks
Stric
it's an lsi-1030
axisys
Stric: is it any good?
Stric
4 connections, of which it can do raid1 on 2 of them
so pretty basic
peerce
the 1030 is a good SAS controller but a lame raid controller, has rather limited cache, and no battery by default, although SOME cards add
Stric
umm.. v440 is not sas.. it's u320 scsi, sca..
peerce
i didn't think the v440 even had a raid controller, just scsi jbod ?
at least for its internal drives
Stric
as I said, it can do 1..4 disk jbod or 2 disk raid1 + 2 jbod
peerce
we've gotten rather mixed results from our database benchmarking on zfs. if we're running oracle hard, we tend to use 80% of CPU while doing an almost pure write load on a large number of SAN volumes. ZFS takes a fair amount of CPU away from oracle
jamesd2
peerce, do you have compression enabled? did you set recordsize to 8k? did oracle performance dispote its excessive cpu usage?
peerce
we did all the recommended sun and oracle zfs tuning
our CPU usage in oracle is because we have a ton of business logic in pl/sql
its the nature of our backend app
axisys
Stric: here is a t5140/5240 with lsi 1068E http://blogs.sun.com/relling/entry/ras_in_the_t5140_and
Stric: is that a good raid controller ?
jamesd2
peerce, it may be wise to add a cpu to the box, but not allow oracle to use it.. could give you a performance boost without the oracle tax.
Stric
1068e in t5120 as well.. haven't used it for anything but jbod
axisys
Stric: jbod? w/ software raid ?
Stric
yes
all 2 disks
axisys
Stric: heh
MilkmanDan
jamesd2: The problem is, Oracle doesn't like you doing that. They say they'll accept Containerization of the database for the purposes of determining CPU/core count, but then they say "oh! Not running in the global Zone! UNSUPPORTED!"
They're being greedy, just like they always have.
Otherwise, yeah, it'd make perfect sense to dedicate a few cores to the OS so ZFS could stay well fed, and shave that off the DB cost of the box.
AJC_Z0
This puts Google in the position of not supporting their product running on their hardware in the way it's designed to be used
MilkmanDan
s/Google/Oracle/
And yes, now that they own Solaris, that's precisely correct.
But since they've never been a company to allow loopholes for customers to eke out more performance than they were paying for, expect everything to change once you finally get a handle on it.
Repeat, ad nauseum.
AJC_Z0
(Action) was dealing with the liars, thieves and charlatans at Google
zetan
any know ce.conf configuration details?
i'm trying to enable jumbo frames on my 4 ceX devices and the settings dont 'take'
peerce
that reminds me, I have to figure that out myself :D
zetan
name="pci100b,35" parent="/pci@8,700000" unit-address=0 accept-jumbo=1;
reboot, and ifconfig ce0 mtu 9194 fails
peerce
huh.
no idea
jamesd2
does it support mtu's that high i thought the most commonly supported jump packet was 9000
kop_
it isn't about morals it's about not buying a car with the hood welded shut
it's about being independent of the mindless addiction and irrational upgrade path of a monolithic ,monotheist monopoly
it's about reasoning for ones self even if it includes M$hit products
choice now where were we
Oh yeah w7 sux dawg bawlls
zetan
9194 was the value from the gigaswift manual (which is ripe with typos and spelling errors)
peerce
kop; so you have the source to your car's EFI processor ?
jamesd2
i would try 9000 and perhaps even 8000-8500 just to be sure that you aren't exceeding some internal limit
peerce
might as well be welded shut, there's nothing a regular person can do under there other than check the oil and water
zetan
which is why I stick with mechanically controlled vehicles
(Action) points to throttle cable and hydraulic brake actuators
no computer at all
peerce
zetan; so nothing built since the 70s ?
zetan
not true
1984 GMC 6.2L diesel G-van
kop_
peerce, actually yes but that's a bad example in my case
zetan
and my 1995 Chevy 6.5L diesel G-van
peerce
mechnical diesel injection on those?
zetan
the 'computer' does very little in these cases. mostly dash lights, trouble codes, and transmission stuff
jamesd2
kop is not normal has has more 8 and 16bit bit microcontrollers in his house and garage than can be found anywhere on earth ;-)
zetan
yeah both
peerce
that must be efficient.
zetan
20mpg freeway with a big load
peerce
nice big cloud of black smoke when you step on it?
kop_
peerce, my own aftermarket closed loop
peerce
kop; so not street legal?
zetan
yeah I can make a smoke show on demand, but usually it's clean
kop_
but I do own a 94 exploder
and I supose in that case it's valid
zetan
I'm hoping to add hydroxy injection (HHO) soon to solve the smoke issue
peerce
hho, you mean, water.
kop_
peerce, meets emissions so who phuckin cares
peerce
i guess you don't have smog inspections where you live
zetan
well, splitting water into 2 H and an O and injecting that gas into the air intake
it's well documented it proves milage and reduces emissions
not a fuel replacement, mind you
kop_
peerce, they don't inspect , yet, alll they do is sniff
peerce
and how do you do this split?
electrolysis ?
zetan
yeah
using about 10amps off the alternator
PWM controlled
KOH as a catalyst
and a bubble to filer the KOH vapor
i have a proof of concept device working
kop_
humm
Ever heard of Stanlel Meyer ?
zetan
i hesitate to say "search youtube for HHO" because of all the redneck versions of this tech
kop_
Ever heard of Stanley Meyer ?
peerce
yah, lotta snakeoil out there and BS claims (K&N Filters <cough>)
kop_
look on youtube for "it runs on water"
zetan
i'm on the hydroxy yahoo group. lots of good research flying around
yeah you spend more energy splitting that you get in return
peerce
I'd rather read an engineering article that had been peer rewviewed, frankly.
zetan
i know ,it's a bunch of numb nuts doing it
but there is some validity in the work
kop_
peerce, many people have looked into Meyer's snake oil
peerce
i just hate watching 'talking heads' videos
kop_
peerce, I'll do some editing and drop it on ya
you're a bright guy , you don't need to crawl before you run
it's in a down'd box atm
peerce, save you the trouble of trolling rexweb and patents online to sift through all the mysticism and smoke show salesmen
the basic principal is interesting
peerce
i don't have audio right where i am atm anyways, on a laptop, using remote control
and video over remote is wretched
gws
how do I find otu if solaris 10 is seeing my emulex fiber card? a hitchi guy can see it in the SAN, but I must now learn how the os sees the space....
:)
kop_
peerce, it won't be any time soon but I'll remember
peerce
gws: otu ?
g4lt-mordant
peerce, typoese...
peerce
gws; assuming everything is setup, to see new LUNs that are mapped to your host, use `devfsadm -v`
gws
devfsadm -v returns nothing, I think the Luns have yet to be assigned...
I am so weak at SAN
zetan
anything show up in 'format' ?
i'm using qlogic cards, and those tools show you the SAN network, can browse luns, do diags, etc. i dont know the emulex variety
SANsurfer!
gws
format?
format shows only disk 0
time to call the party people at sun support!
:)
I have to whip people here to make em use raid 0+1
lol
they want performance
our dba a joker
WOUniversity
Hello all, I was hoping someone here could help me.
I am trying to integrate some Solaris machines into my Active Directory hosted by a Windows 2008 Server.
They are running Solaris 10, I need to somehow enable Kerberos and am currently at a loss for what to do.
Has anybody here seen or done anything like that?
peerce
are the windows servers using the Microsoft Unix extensions?
oh, he left.
zetan
some of our customers use libVas for AD integration
but sadly libvas is not thread safe
doubletwist
I dunno, but I suspect we will be looking into authenticating against the AD servers soon. Within a year perhaps.
peerce
as I understand it, and I've not set this up, if you use Microsoft's Unix extensions, its pretty straight forward, you configure unix for LDAP
you have to populate the unix user's AD records with the unix UID/GID you want them to use
doubletwist
right
I'm not looking forward to it. I've tried on my own playing around at home to get my linux boxes to authenticate against an openldap server. I came to the conclusion that I'm too stupid to understand ldap :)
Chubbz
Does someone know of a good resource on Solaris hard drive notation?
peerce
um, cXtYdZsW, where X is the channel, Y is the target, Z is the device or lun# on that target, and W is the slice, 0-9