Boot Disk on an EMC disk Array, How do you like it ?

[R.M.PANDIAN]

Hi Fello H-puxer,

Some of my peers feel it's a good idea to use an EMC volume(split) as the BOOT disk.
(All disks in one Black box policy!!!)
(I don't like it, but I can't convince them)

Apart from the fact that we must make sure that the PDC should be able to see the volume on the EMC array and there may be some F/W upgrade and Host Bus Adaptors (SCSI or F/C)purchase required, what do you feel about the concept of booting from EMC Disk Array???

Thanks in advance,
Pandian.

[Tommy Palo]

I defenitely don't like it !
An EMC array should in my opinion be used for storage, and not be cluttered with all kinds of operating systems.
Boot/OS disks should be local and if redundancy is the issue use two disks and MirrorDisk/UX for that purpose.

[Alexander M. Ermes]

I do not like the idea either.
We are using a V-class as database server
running 21 databases. We have configured
a High Availability Storage System with the boot system. We use the EMC only for the databases. Reason : if the EMC goes down, you can still operate your server. If you connected the EMC with fiber channel, there is no chance anyway.
Rgds
Alexander M. Ermes

[James R. Ferguson]

Hi, Pandian:

Interestingly this discussion surfaced recently in my shop. My "gut" feeling is that I don't like this either, but this is entirely subjective at this point. My feelings are that I want to be able to boot a server as a self-contained unit, divorced from as many external devices as possible for simplicity, the most flexability and fault isolation.

I suppose one argument in favor of intelligent storage arrays, like EMC2, is that their large I/O cache can offer great performance gains.

I always run mirrored boot disks, and so, argue that the high-availabiity offered by products like the EMC2 Symmetrix are not an issue with boot devices.

I firmly believe in the tenet that only the operating system belongs on a boot disk. We use an EMC2 array for our Oracle databases and application code on two K570 servers and on a Unisys mainframe and its database and applications. I have everything positive to say about that environment, but intrinsicly hold to being able to boot a self-contained server.

...JRF...

[Bill Hassell]

Booting isn't the only issue here. All your vg00 traffic is going over the same channel (unless you've setup additional channels to the EMC). I have personally moved 3 customers off their arrays back to internal disks where performance is *much* better.

Putting all your eggs in one basket means that when the EMC melts down, all your machines are useless. With internal disks, you can at least continue running and perhaps switch some network resource for backup data.

I've seen a dozen PC servers and 6 HP-UX boxes connected to a big SAN as their only disk...the entire site went down hard when the SAN stopped working. And to add insult to injury, someone had pulled out the phone cable so the SAN box could phone home. (Apparently the monitoring site does not poll) No one knew what went wrong but most of a weekend's production was gone and there was no way to boot any of the machines.

[paul courry]

I'm glad everyone is stipulating HP-UX because on the HP3000 side of the house EMC is the best of a bad lot of choices. There is no mirroring for ldev 1, the model 20 is old and requires the expensive 620 processor, Autoraid? I don't know. We have had good reports from people who have done it on 3000's. As far as the arguments about keeping the server running if the EMC goes down, if all your data is on the EMC just what are your programs going to process if it goes down?

[Patrick Wallek]

I agree with everyone else that thinks having boot disks on a RAID Array is a very BAD idea.

I also have all of my OS on HASS units and I use mirroring and have all of vg00 mirrored.

As far as the question about the system being up if the array fails..... Even if the disk array fails, I want to have the system up so I can prove that the problem is not with the OS, but with the hardware. If you can not get into the system, then you have no hope of running any diagnostics if something happens to fail.

If you have mirrored OS drives your chances are almost 0 that your system will crash because of the loss of an OS drive.

[Rita C Workman]

I didn't read all the responses so far....BUT I can say definitely NO...
When I got here...a previous person decided to extend over to our EMC disk array. I didn't like and mentioned it to mgmt..who felt it's been ok and the other guy was such a 'whiz'. About 7 months later, hardware problem hit. Seems the fiber card went bad to the EMC. The EMC was fine...but the box wasn't. So here's the scenario....when you lose (for any reason) your connection to the EMC you lose your 'small' control of your operating system. You know those unimportant things like getting to single user mode to fix something, or the ability to boot off the mirror. Because they are all out of your reach on the EMC.......arghhhhh!!
It was a nightmare, I never want to live through again.
Needless to say, the purse strings opened back up and the O/S was put back on the drives it belonged.

If you have your boot drive mirrored....you can get back up and stay up on that until the primary is replaced.

So .... my vote (hope this is counted in Florida) is 'nay'

rcw

[R.M.PANDIAN]

Thank you very much for all the replies.

Just to summarise the replies so far.

1) Everyone thinks it's a bad idea.
2) If the HBA (SCSI or F/C) is shared by data and O/S, there may be performace problems.
3) Problem determination is complicated, if the problem is with the array or HBA connecting the host to the array or the SAN as you won't have the O/S up and running.
4) If your array is down, not only the data but also the O/S is unavailable.

Just playing devil's advocate,

Bill,

Can you explain the 3 customers scenarios bit more.

i) Was it a SCSI or F/C direct connect or SAN ?
ii) What if they used dedicated HBA for boot volume only?
iii) About SAN, what if one chooses to use only the direct connect array for booting (by pass the SAN for boot only)?

Finally, the maximum thorughput of the internal SCSI is 40 MB/s, but a well configured root disk connected though F/C can deliver upto 100 MB/s (under ideal circumstances). So it perform better.

Rita,

Could you have avoided the problem if you only used a redundant HBA. (which will be the way to go for even internal root disk for production systems)

Paul C brings in an argument that if the data is unavailable, what's the use of host being up and running.


As a traditional H-puxer, I still believe it's not a good idea. But, i think we have to have more explanation on why it is?

Your time and effort will be much appreciated.

Thanks
Pandian.

[Dan Hetzel]

Hi,

I used to work in France, UK and now Belgium with EMC storage units of all sizes.

One point is absolutely constant, EMC people from those different countries agree that booting from EMC should be avoided.
I've asked that question many times, including last week when we ordered a complete Connectrix SAN solution. The answer was always: "you shouldn't..."

I personally prefer to have the ability to boot my servers in standalone mode.
I agree that a F/C is a lot faster than internal SCSI but, depending on the number of systems connected, you could use a good part of the F/C bandwith just to run unix and the utilities.

Best regards,

Dan

[Rick Garland]

Like everyone else, my opinion in a resounding "DON'T LIKE IT".

For me, boot disks and OS disks are in a Jamica style setup with mirroring. If a disk fails (you know it will happen sooner or later), you have a hot-plug disk ready to go and you are back. This process takes what, some 10 - 20 minutes. Ask EMC how long it would take for a replacement disk.

For storage, EMC units are good. For keeping the box w/ OS running, EMC is not the preferred choice.