|
Home > Archive > Data Storage > January 2005 > Can iSCSI work in 24/7 environment?
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
Can iSCSI work in 24/7 environment?
|
|
| Rob Scott 2004-07-28, 5:45 pm |
| I am looking at using consolidated storage to ease our fragmented and
increasingly hard to manage storage problem at my company. The conventional
way to go seems to be to buy an expensive fibre channel SAN from HP or EMC,
buy FC cards for all the servers, dual FC switches, and an FC array with
dual redundant FC-SCSI controllers.
But iSCSI keeps popping up as a cheaper alternative what with using exsiting
giga infrastructure. I have done some tests with Linux and MS clients
connecting to simple a Linux target but there seem to be problems with using
this kind of thing operationally. The biggest one I find is redundancy.
Is it possible to make a fully redundant iSCSI disk storage array?
You can get SATA storage arrays on the market but they usually have a SCSI
connection and are not dual redundant. The SATA controller becomes a risk.
But even if that weren't the case, the array connector should somehow
connect to two server controllers to provide redundancy.
In short, I can see how to make an iSCSI storage array, put it on the
network, and get initiators to connect to it, but there would be very little
redundancy. There are several single points of failure on the storage array
which would preclude it from 24/7 mission critical use.
Anyone know of a way around this? Or am I going to have to wait for bespoke
iSCSI products to come to market?
| |
| Jesper Monsted 2004-07-28, 5:45 pm |
| "Rob Scott" <rob.scott@fathomtechnology.com> wrote in
news:41066347@andromeda.datanet.hu:
> In short, I can see how to make an iSCSI storage array, put it on the
> network, and get initiators to connect to it, but there would be very
> little redundancy. There are several single points of failure on the
> storage array which would preclude it from 24/7 mission critical use.
>
> Anyone know of a way around this? Or am I going to have to wait for
> bespoke iSCSI products to come to market?
I'm not sure if it's out yet, but EMC are doing iSCSI targets on their
Symmetrix high-end storage. The network itself would have to be made
redundant, but how the iscsi initiators would use it, i don't know.
--
/Jesper Monsted
| |
| Faeandar 2004-07-28, 5:45 pm |
| On Tue, 27 Jul 2004 16:14:28 +0200, "Rob Scott"
<rob.scott@fathomtechnology.com> wrote:
>I am looking at using consolidated storage to ease our fragmented and
>increasingly hard to manage storage problem at my company. The conventional
>way to go seems to be to buy an expensive fibre channel SAN from HP or EMC,
>buy FC cards for all the servers, dual FC switches, and an FC array with
>dual redundant FC-SCSI controllers.
>
>But iSCSI keeps popping up as a cheaper alternative what with using exsiting
>giga infrastructure. I have done some tests with Linux and MS clients
>connecting to simple a Linux target but there seem to be problems with using
>this kind of thing operationally. The biggest one I find is redundancy.
>
>Is it possible to make a fully redundant iSCSI disk storage array?
>You can get SATA storage arrays on the market but they usually have a SCSI
>connection and are not dual redundant. The SATA controller becomes a risk.
>But even if that weren't the case, the array connector should somehow
>connect to two server controllers to provide redundancy.
>
>In short, I can see how to make an iSCSI storage array, put it on the
>network, and get initiators to connect to it, but there would be very little
>redundancy. There are several single points of failure on the storage array
>which would preclude it from 24/7 mission critical use.
>
>Anyone know of a way around this? Or am I going to have to wait for bespoke
>iSCSI products to come to market?
>
An iSCSI environment can be made just as redundant as a FC SAN
environment. I can personally say that all aspects work with the
exception of the iSCSI failover. We only briefly tested it on NetApp
and it appeared not to work. Granted it was an ad hoc test with an
uncertain application but still. I just can't vouch for it
personally.
~F
| |
| Faeandar 2004-07-28, 5:45 pm |
| On Tue, 27 Jul 2004 22:34:26 GMT, Faeandar <mr_castalot@yahoo.com>
wrote:
>On Tue, 27 Jul 2004 16:14:28 +0200, "Rob Scott"
><rob.scott@fathomtechnology.com> wrote:
>
>
>An iSCSI environment can be made just as redundant as a FC SAN
>environment. I can personally say that all aspects work with the
>exception of the iSCSI failover. We only briefly tested it on NetApp
>and it appeared not to work. Granted it was an ad hoc test with an
>uncertain application but still. I just can't vouch for it
>personally.
>
>~F
I feel remiss in not mentioning the other pieces since the iSCSI
failover part is fairly significant... 
All the networking pieces can be made redundant. Alot of people don't
realize that ethernet can be made as fail-proof as FC SAN's, it's just
that generally the size of an ethernet network is so much larger that
it's not cost justified in most cases.
However, NetApp iSCSI is *supposed* to failover fine, just like the
other protocols it handles. It's even supposed to failover to a filer
that does not have an iSCSI license..
~F
| |
| ahedge 2004-07-28, 8:45 pm |
|
"Rob Scott" <rob.scott@fathomtechnology.com> wrote in message
news:41066347@andromeda.datanet.hu...
> I am looking at using consolidated storage to ease our fragmented and
> increasingly hard to manage storage problem at my company. The
conventional
> way to go seems to be to buy an expensive fibre channel SAN from HP or
EMC,
> buy FC cards for all the servers, dual FC switches, and an FC array with
> dual redundant FC-SCSI controllers.
>
> But iSCSI keeps popping up as a cheaper alternative what with using
exsiting
> giga infrastructure. I have done some tests with Linux and MS clients
> connecting to simple a Linux target but there seem to be problems with
using
> this kind of thing operationally. The biggest one I find is redundancy.
>
> Is it possible to make a fully redundant iSCSI disk storage array?
> You can get SATA storage arrays on the market but they usually have a SCSI
> connection and are not dual redundant. The SATA controller becomes a risk.
> But even if that weren't the case, the array connector should somehow
> connect to two server controllers to provide redundancy.
>
> In short, I can see how to make an iSCSI storage array, put it on the
> network, and get initiators to connect to it, but there would be very
little
> redundancy. There are several single points of failure on the storage
array
> which would preclude it from 24/7 mission critical use.
>
> Anyone know of a way around this? Or am I going to have to wait for
bespoke
> iSCSI products to come to market?
>
>
Have you considered Equallogic and Left Hand Networks? There are probably
other names that I am missing. Also NetApp FAS2xx should be worth a look.
iSCSI boxes with redundancy and failover are expensive, but you could still
save some money on HBAs and switches.
Also, if you decide for FC look beyond HP and EMC. Xiotech?
Good luck
Art
| |
| Rob Scott 2004-07-29, 7:45 am |
| Faeandar wrote:
> On Tue, 27 Jul 2004 16:14:28 +0200, "Rob Scott"
> <rob.scott@fathomtechnology.com> wrote:
>
>
>
>
> An iSCSI environment can be made just as redundant as a FC SAN
> environment. I can personally say that all aspects work with the
> exception of the iSCSI failover. We only briefly tested it on NetApp
> and it appeared not to work. Granted it was an ad hoc test with an
> uncertain application but still. I just can't vouch for it
> personally.
>
> ~F
It is clear that the network components can easily do failover and
redundancy and so on. That is just down to simple networking.
The problem is what is attached to the network down to the disk drives.
If you just have a bunch of disks in a PC, even if you have a RAID array
for the disks, then the motherboard is a single point of failure.
You could have an external disk array connected via SCSI to two PCs in a
failover cluster confugeration.
Or you could put two identical iSCSI boxes on the net and RAID between them.
None of these solutions is at all ideal.
What you want to end up with is a SATA disk array, with redundant SATA
controllers (not seen it so far) and two ethernet ports. The disk can be
RAID anything you like.
I don't see anything like this on the market yet. The simple solution
gives you no redundancy past the ethernet port to the PC. Solutions with
dual failover clustered PCs or RAID over 2 iSCSI devices are either
tricky to implement or expensive (RAID 1 costs you double) or tricky to
administer (in the case of RAID over iSCSI).
| |
| Rob Scott 2004-07-29, 7:45 am |
| ahedge wrote:
> "Rob Scott" <rob.scott@fathomtechnology.com> wrote in message
> news:41066347@andromeda.datanet.hu...
>
>
> conventional
>
>
> EMC,
>
>
> exsiting
>
>
> using
>
>
> little
>
>
> array
>
>
> bespoke
>
>
>
> Have you considered Equallogic and Left Hand Networks? There are probably
> other names that I am missing. Also NetApp FAS2xx should be worth a look.
> iSCSI boxes with redundancy and failover are expensive, but you could still
> save some money on HBAs and switches.
>
> Also, if you decide for FC look beyond HP and EMC. Xiotech?
>
> Good luck
>
> Art
>
>
It looks to me that the LHN products do not have redundancy beyond
having dual NICs.
The Equallogic products may have.
Still looks like iSCSI has a way to go to be used for anything other
than non-critical storage. I wouldn't want to run critical email,
database, and other things on it if there wasn't dual redundancy.
If the thing went down because for example the processor board failed,
then the entire company would be brought to its knees if all the servers
got their storage from there.
Thanks for the pointers.
| |
| Benno... 2004-07-29, 8:05 am |
| Rob Scott wrote:
> It is clear that the network components can easily do failover and
> redundancy and so on. That is just down to simple networking.
>
> The problem is what is attached to the network down to the disk drives.
> If you just have a bunch of disks in a PC, even if you have a RAID array
> for the disks, then the motherboard is a single point of failure.
>
> You could have an external disk array connected via SCSI to two PCs in a
> failover cluster confugeration.
>
> Or you could put two identical iSCSI boxes on the net and RAID between
> them.
>
> None of these solutions is at all ideal.
>
> What you want to end up with is a SATA disk array, with redundant SATA
> controllers (not seen it so far) and two ethernet ports. The disk can be
> RAID anything you like.
Does iSCSI mean your SAN box *must* use SCSI (or SATA) disks? If your
SAN box uses FC disks it can be made fully redundant right up to the
disks themselves, right? Your SAN box will still be very expensive but
your connections will be much cheaper when done with iSCSI over Gigabit
Ethernet.
B.
| |
| Bill Todd 2004-07-29, 5:45 pm |
|
"Rob Scott" <rob.scott@fathomtechnology.com> wrote in message
news:41066347@andromeda.datanet.hu...
....
> Is it possible to make a fully redundant iSCSI disk storage array?
I guess that the short answer would have to be 'no', unless you can find
SATA drives that have two connectors on them as high-end FC drives do:
otherwise, drive-interconnect or controller failure will always create
apparent drive failure in a manner that is not a problem in fully-redundant
FC storage.
On the other hand, if that can be discounted as a drawback a qualified
answer may be 'yes': just use two completely redundant paths from host(s)
to separate iSCSI SATA arrays, and mirror between them. Unfortunately, this
doesn't give you the kind of no-single-point-of-failure *box* that you can
purchase from people like EMC: it requires the fail-over and mirroring
intelligence to reside in the host(s).
I'd think it would be possible to build an iSCSI box with EMC-style
redundancy at least up to the back-end disk controllers (and possibly even
to the final links to the disks, if SATA disks attached via SAS turn out to
be able to handle dual initiators), but (just as is the case with existing
FC fully-redundant boxes) some vendor would have to create such a
proprietary combination out of otherwise standard components.
- bill
| |
| ahedge 2004-07-29, 5:45 pm |
|
"Rob Scott" <rob_scott@epam.com> wrote in message
news:4108fa40@andromeda.datanet.hu...
> ahedge wrote:
>
SCSI[vbcol=seagreen]
risk.[vbcol=seagreen]
probably[vbcol=seagreen]
look.[vbcol=seagreen]
still[vbcol=seagreen]
>
> It looks to me that the LHN products do not have redundancy beyond
> having dual NICs.
> The Equallogic products may have.
>
> Still looks like iSCSI has a way to go to be used for anything other
> than non-critical storage. I wouldn't want to run critical email,
> database, and other things on it if there wasn't dual redundancy.
> If the thing went down because for example the processor board failed,
> then the entire company would be brought to its knees if all the servers
> got their storage from there.
I hear you, but I don't follow on the iSCSI-may-not-be-for-serious-business
path. I would not bet my company on a possibly failing board either.Also I
would not take chances on any single array of disks, no matter how many
controllers are in it.
In your initial post you said: "I am looking at using consolidated storage
to ease our fragmented and increasingly hard to manage storage problem at my
company" .
I hear the pain, "increasingly hard to manage" which is a good reason to
move away from what you have, but the "consolidated storage" part seems to
indicate that you want to put everything on a single array. Well, if that's
the case, don't: it's a rather bad idea, with iSCSI, FC or even plain SCSI.
Perhaps the answer is building your consolidated storage layout with an eye
on failover (BC, DR, whatever..)? We know that those things are going to
fail, so why not design a failproof network in the first place. After
that, choosing the pieces to build it should be a trivial matter of
measuring the bang for the buck from each vendor/technology/product.
> Thanks for the pointers.
You're welcome, but I should have mentioned also Falconstor and Datacore,
two software only shops worth considering, especially if your company has
already spent beaucoup money on hardware.
| |
| Anton Kolomyeytsev 2004-07-29, 5:45 pm |
| The power of iSCSI is virtualization. You should not know what type of
storage is used by iSCSI target to allocate the LUN from it provides
you with. It can be JBOD created from el cheapo SATA disks and
exported to you as single huge volume, existing FC array with software
RAID or unallocated sparse file places on NTFS or whatever file
system...
Regards,
Anton Kolomyeytsev
CEO, Rocket Division Software
"Benno..." <0@0.invalid> wrote in message news:<4109039c$0$273$4d4ebb8e@news.nl.uu.net>...
> Rob Scott wrote:
>
>
> Does iSCSI mean your SAN box *must* use SCSI (or SATA) disks? If your
> SAN box uses FC disks it can be made fully redundant right up to the
> disks themselves, right? Your SAN box will still be very expensive but
> your connections will be much cheaper when done with iSCSI over Gigabit
> Ethernet.
>
> B.
| |
| Jesper Monsted 2004-07-29, 5:45 pm |
| "Benno..." <0@0.invalid> wrote in
news:4109039c$0$273$4d4ebb8e@news.nl.uu.net:
> Does iSCSI mean your SAN box *must* use SCSI (or SATA) disks? If your
> SAN box uses FC disks it can be made fully redundant right up to the
> disks themselves, right? Your SAN box will still be very expensive but
> your connections will be much cheaper when done with iSCSI over
> Gigabit Ethernet.
Not at all. iSCSI can be used with any type of disk you fancy. EMC's
Symmetrix has an iSCSI director blade that'll give you a fully redundant
storage system with redundant iSCSI interface ports (GigE). Add some
decently configured initiator software that'll do failover and you're all
set.
--
/Jesper Monsted
| |
|
| On 29 Jul 2004 15:00:39 -0700, anton@rocketdivision.com (Anton
Kolomyeytsev) wrote:
>The power of iSCSI is virtualization. You should not know what type of
>storage is used by iSCSI target to allocate the LUN from it provides
>you with. It can be JBOD created from el cheapo SATA disks and
>exported to you as single huge volume, existing FC array with software
>RAID or unallocated sparse file places on NTFS or whatever file
>system...
>
How is this "power of ISCI" different than the solution of direct FC
connection for the server?
A FC-connected server also gets "virtualization" from the standpoint that
the server doesn't know what's on the other end of the LUN. Heck, for that
matter, a typical SCSI RAID controller hides that from the OS.
I note that the only thing missing is the ability to present LUNs from NTFS
file systems... but not sure that's all that cool a feature anyway.
--- jls
The preceding message was personal opinion only.
I do not speak in any authorized capacity for anyone,
and certainly not my employer.
(get rid of the xxxz in my address to e-mail)
| |
| Anton Kolomyeytsev 2004-07-31, 7:45 am |
| 1) No big difference. However in case of FC you can plug FC device
directly to FC controlller (virtualization layer is kind of optional),
and in case of iSCSI it's not physically possible so virtualization
layer is simply assumed.
2) Yes, kind of... I cannot imagine RAID controller creating resulting
volume set from anything except hard disks. However in case of iSCSI
target emulated SCSI hard disk can have any back end media (true SCSI
hard disk mapped AS IS, file on file system, SATA RAID volume set,
snapshot etc etc etc). Just a bit wider choice.
Regards,
Anton Kolomyetsev
CEO, Rocket Division Software
jlsue <jefflsxxxz@sbcglobal.net> wrote in message news:<0iokg09s042ofcmkmdki3rf114r5n5h4iq@4ax.com>...
> On 29 Jul 2004 15:00:39 -0700, anton@rocketdivision.com (Anton
> Kolomyeytsev) wrote:
>
>
> How is this "power of ISCI" different than the solution of direct FC
> connection for the server?
>
> A FC-connected server also gets "virtualization" from the standpoint that
> the server doesn't know what's on the other end of the LUN. Heck, for that
> matter, a typical SCSI RAID controller hides that from the OS.
>
> I note that the only thing missing is the ability to present LUNs from NTFS
> file systems... but not sure that's all that cool a feature anyway.
>
> --- jls
> The preceding message was personal opinion only.
> I do not speak in any authorized capacity for anyone,
> and certainly not my employer.
> (get rid of the xxxz in my address to e-mail)
| |
| Nik Simpson 2004-07-31, 7:45 am |
| Anton Kolomyeytsev wrote:
> 1) No big difference. However in case of FC you can plug FC device
> directly to FC controlller (virtualization layer is kind of optional),
> and in case of iSCSI it's not physically possible so virtualization
> layer is simply assumed.
>
> 2) Yes, kind of... I cannot imagine RAID controller creating resulting
> volume set from anything except hard disks. However in case of iSCSI
> target emulated SCSI hard disk can have any back end media (true SCSI
> hard disk mapped AS IS, file on file system, SATA RAID volume set,
> snapshot etc etc etc). Just a bit wider choice.
But none of these are features of iSCSI per-se, iSCSI is a block level I/O
transport just like SCSI, SATA, FC, etc, there is no requirement for
implementations of iSCSI targets to anything more than pass blocks to disk
drives on the back end. All the things you've talked about have implemented
in one place or another on top of a variety of block level I/O protocols.
That you are seeing them in a lot of early iSCSI implementations is becuase
these functions are now accepted as desirable and people designing a new
generation of I/O controllers are implementing them from day one rather than
trying to bolth them on as an after thought. But iSCSI is not the reason for
this, without iSCSI you would be seeing these functions appear on other
types of controllers, in fact you are seeing these things appear on other
types of controllers.
--
Nik Simpson
| |
|
| In article <82MOc.6439$g_5.5083@bignews4.bellsouth.net>,
n_simpson@bellsouth.net says...
>But none of these are features of iSCSI per-se, iSCSI is a block level I/O
>transport just like SCSI, SATA, FC, etc, there is no requirement for
>implementations of iSCSI targets to anything more than pass blocks to disk
>drives on the back end. All the things you've talked about have implemented
>in one place or another on top of a variety of block level I/O protocols.
>That you are seeing them in a lot of early iSCSI implementations is becuase
>these functions are now accepted as desirable and people designing a new
>generation of I/O controllers are implementing them from day one rather
than
>trying to bolth them on as an after thought. But iSCSI is not the reason
for
>this, without iSCSI you would be seeing these functions appear on other
>types of controllers, in fact you are seeing these things appear on other
>types of controllers.
>
exactly
Adaptec has just started delivering their isa1500 (1TB / 750GB usable)
SATA iSCSI storage array WITH a special package including 2 units
& their replication software for around $10k
we'll have a live internet based demo program going within the month
with one of their boxes attached to one of our web servers that will
allow people to connect directly to some "live" iSCSI storage & test
with it
_____ . .
' \\ . . |>>
O// . . |
\_\ . . |
| | . . . |
/ | . www.EvenEnterprises.com . . . |
/ .| aseltzer@EvenEnterprises.com . . . |
/ . | 310-544-9439 / 310-544-9309 fax . . . o
----------------------------------------------------------------------
Authorized - DIRECT VAR/VAD/Distributor for new SCSI/FC-AL peripherals
NAS/SAN/RAID from HP, Seagate, EMC, Adaptec, Qlogic, OverLand Storage
| |
| Rob Scott 2004-08-06, 5:45 pm |
| Yep I can imagine having two boxes in a mirror arrangement, but then you
have just doubled your costs and so have diminished your advantage
against FC SANs.
It seems that there is no dual redundant all-in-the-box arrangement with
two of everything in it yet, if ever.
One non-redundant box is OK for non-critical stuff, like disk-to-disk
backup, or even email (as long as there is certain types of redundancy
so you don't lose the data just have some downtime), and non-critical
file sharing stuff. But for say source control systems (we are a coding
shop) we couldn't afford downtime over all our projects at the same
time. Currently we have a server per project with box attached storage
so it is unlikely that all projects are down at the same time.
I am interested to see how these iSCSI products develop and get more
sophisticated or whether they just stay at the disk-to-disk backup level.
Bill Todd wrote:
> "Rob Scott" <rob.scott@fathomtechnology.com> wrote in message
> news:41066347@andromeda.datanet.hu...
>
> ...
>
>
>
>
> I guess that the short answer would have to be 'no', unless you can find
> SATA drives that have two connectors on them as high-end FC drives do:
> otherwise, drive-interconnect or controller failure will always create
> apparent drive failure in a manner that is not a problem in fully-redundant
> FC storage.
>
> On the other hand, if that can be discounted as a drawback a qualified
> answer may be 'yes': just use two completely redundant paths from host(s)
> to separate iSCSI SATA arrays, and mirror between them. Unfortunately, this
> doesn't give you the kind of no-single-point-of-failure *box* that you can
> purchase from people like EMC: it requires the fail-over and mirroring
> intelligence to reside in the host(s).
>
> I'd think it would be possible to build an iSCSI box with EMC-style
> redundancy at least up to the back-end disk controllers (and possibly even
> to the final links to the disks, if SATA disks attached via SAS turn out to
> be able to handle dual initiators), but (just as is the case with existing
> FC fully-redundant boxes) some vendor would have to create such a
> proprietary combination out of otherwise standard components.
>
> - bill
>
>
>
| |
| Bill Todd 2004-08-06, 8:45 pm |
|
"Rob Scott" <rob_scott@epam.com> wrote in message
news:4113c353@andromeda.datanet.hu...
> Yep I can imagine having two boxes in a mirror arrangement, but then you
> have just doubled your costs and so have diminished your advantage
> against FC SANs.
>
> It seems that there is no dual redundant all-in-the-box arrangement with
> two of everything in it yet, if ever.
Er, mirroring two simple boxes is likely to cost you a good deal *less* than
that single everything-redundant box you're so interested in. In fact, it's
likely the lowest-cost redundant solution you can build, save for
direct-attached storage.
- bill
| |
| monikeo@gmail.com 2005-01-07, 5:46 pm |
| I"m using the Equallogic system with replication between two sites- LA
& SF. That gives me the reduncy that I need to restore quickly from
either site. The only problem is the replicas takes up 200% of the LUN.
|
|
|
|
|