iSCSI clustering

Lrrpie-CT · 09-23-2010, 12:26 AM

can kiss my @ss. That is all. Good night.

criticalhammer · 09-23-2010, 12:56 AM

I used to

administrating a EMC SAN using iSCSI. Fibercards and GBIC modules would blow all the time and yet the SAN wouldn't recognize what happened. I would have to rebuild connections from scratch.

tarakian · 09-23-2010, 09:19 AM

Just set one of those up last week. The cluster part wasn't bad. Setting up an SQL cluster on top of it was ridiculous.

Lrrpie-CT · 09-23-2010, 09:29 AM

After six hours of sleep I'm mostly refreshed. It is the "High Availability" aspect that was driving me nuts. Basically a RAID1 of two iSCSI targets so if one fails the other takes over with no down time. They are supposed to sync in the background. Works great even after pulling the plug on one in a simulated failure. I was very proud of myself. But, when the "RAID1"-like partner comes back online they need to sync again. This takes both iSCSI targets offline during syncing. Downtime can be hours. Doesn't sound very "High Availability" to me. What a pain in the @ss. Decided to ditch the HA component and figure something else out. Whatever...

criticalhammer · 09-23-2010, 09:47 AM

If you can afford the disks then just mirror whatever is running and have it act like a hot swappable. If the live disks die then the hot swappable disks will pick up where it left off, but don't have it setup for auto rebuild.

BlueDragonX · 09-23-2010, 02:40 PM

Quote:

Originally Posted by criticalhammer

If you can afford the disks then just mirror whatever is running and have it act like a hot swappable. If the live disks die then the hot swappable disks will pick up where it left off, but don't have it setup for auto rebuild.

That would make sense if it was all in one enclosure, but that would defeat the purpose. A single enclosure means a single point of failure. Motherboard dies and you've now got downtime. For real HA you have to consider everything, not just the disks.

Kindred · 09-23-2010, 03:59 PM

Quote:

Originally Posted by BlueDragonX

That would make sense if it was all in one enclosure, but that would defeat the purpose. A single enclosure means a single point of failure. Motherboard dies and you've now got downtime. For real HA you have to consider everything, not just the disks.

Agreed, but unfortunately most of us balance "high availability" and "fault tolerance" with "Practical" and "affordable". (*sigh* I miss the days of getting blank checks for the technology!!)

Since he's using ISCSI we can assume that both storage devices are in the same datacenter. (BIG single point of failure)

Or the cables are run through the same wiring tray for at least part of their journey (smaller risk, but still a single point)

The disks are mirrored, not replicated offsite, so there's no protection against data corruption.

Let's face it, disks are typically the most common point of failure and when they go you can lose data......versus a mother board going, which costs you downtime, but usually not data loss. Even systems that aren't worthy of clustering still get some kind of RAID for their disks.

Lrrpie-CT · 09-23-2010, 04:24 PM

Not that anyone cares too much or asked for that matter but here's what I've got

Server 1:

Dual Quad Core Xeon, 16gb, 2TB RAID 10 array. 4 1gb NIC's. Dual PS(duh), Windows 2008 R2 Enterprise, running Hyper-V, iSCSI target software from Starwind.

Server 2:

Same except one RAID5 array, no RPS.

1 Netgear GS724T gig switch for cluster nodes and LAN/VM traffic
1 3Com 5500G gig switch for iSCSI traffic/heartbeat for the nodes

Both servers are in the same rack with different UPS.

I'm not building NORAD here, just look to build some good fault tolerance for my server farm. So, I can only heap praises on Hyper-V R2 and Live Migration. It kicks @ss! I can have a VM running on node 2, click Live Migration to node 1 and down time is limited to like 2 seconds! So, I want some redundancy in case the iSCSI array goes south. I think that some type of robust backup is the best solution. I've already spent a ton of money on my server farm. As long as the iSCSI SAN is up, cluster nodes running the VM's can be turned on and off/added and removed with impunity with no down time.

Kindred · 09-24-2010, 10:48 AM

Does sound like a very slick setup Matt.
So much so that I won't even start trash talking your HyperV over VmWare decision !

09-23-2010, 12:56 AM	#2 (permalink)
criticalhammer Sly Fox Cyndicate Join Date: Aug 2009 Location: Dark side of the moon	I used to administrating a EMC SAN using iSCSI. Fibercards and GBIC modules would blow all the time and yet the SAN wouldn't recognize what happened. I would have to rebuild connections from scratch. __________________ The American dream, you have to be asleep to believe it. "... it's all about money, ain't a damn thing funny. You got to have a con in this land of milk and honey..." "A considerable percentage of the people we meet on the street are people who are empty inside, that is, they are actually already dead. It is fortunate for us that we do not see and do not know it. If we knew what a number of people are actually dead and what a number of these dead people govern our lives, we should go mad with horror." " Here's a little bit of advice, you're quite welcome, it is free. Donâ€™t do nothing that is cut-price, you'll know what they'll make you be. They will try their tricky device, trap you with the ordinary. Get your teeth into a small slice, the cake of liberty." "Today a young man on acid realized that all matter is merely energy condensed to a slow vibration, that we are all one consciousness experiencing itself subjectively, there is no such thing as death, life is only a dream, and we are the imagination of ourselves." - Bill Hicks R.I.P Bill Hicks

09-23-2010, 09:19 AM	#3 (permalink)
tarakian Stange, but not a stanger Join Date: Mar 2007	Just set one of those up last week. The cluster part wasn't bad. Setting up an SQL cluster on top of it was ridiculous. __________________ No one has the right not to be offended. Feedback www.zombieshoot.com

09-23-2010, 09:47 AM	#5 (permalink)
criticalhammer Sly Fox Cyndicate Join Date: Aug 2009 Location: Dark side of the moon	If you can afford the disks then just mirror whatever is running and have it act like a hot swappable. If the live disks die then the hot swappable disks will pick up where it left off, but don't have it setup for auto rebuild. __________________ The American dream, you have to be asleep to believe it. "... it's all about money, ain't a damn thing funny. You got to have a con in this land of milk and honey..." "A considerable percentage of the people we meet on the street are people who are empty inside, that is, they are actually already dead. It is fortunate for us that we do not see and do not know it. If we knew what a number of people are actually dead and what a number of these dead people govern our lives, we should go mad with horror." " Here's a little bit of advice, you're quite welcome, it is free. Donâ€™t do nothing that is cut-price, you'll know what they'll make you be. They will try their tricky device, trap you with the ordinary. Get your teeth into a small slice, the cake of liberty." "Today a young man on acid realized that all matter is merely energy condensed to a slow vibration, that we are all one consciousness experiencing itself subjectively, there is no such thing as death, life is only a dream, and we are the imagination of ourselves." - Bill Hicks R.I.P Bill Hicks

09-24-2010, 10:48 AM	#9 (permalink)
Kindred Will Code For Schlitz Join Date: Mar 2006 Location: Neshanic Station, NJ	Does sound like a very slick setup Matt. So much so that I won't even start trash talking your HyperV over VmWare decision ! __________________ Strive for that moment when you're only a slice of pizza and a hooker away from paradise.