|
|
|
Forum Newbie
Group: Forum Members
Last Login: Thursday, July 31, 2008 1:19 PM
Posts: 3,
Visits: 15
|
|
I love the concept of VSA. I am using in a test setup without the RAID features. But I am wondering if the diagram I have will work? And about what percentage of native performance I should expect.
I have heard that only 2TB can be managed with VSA? But is it that more than 2TB can be managed but can't be part of the same RAIN volume?
Sorry to ask so many questions, but I thought the best place to get some good answers might be here.
*****Whoops....uploading the diagram didn't work....I hope just listing below would make sense.*****
Many thanks in advance.
(3) Dell 2900's
300 GB X 10 drives, RAID5, ESX3i
VSA1, VSA2, VSA3 respectively
2.7TB usable per unit.
On the "regular" network side I'd have multiple GB NIC's
On the Iscsi Network I'd have 10GbE nics
I'd like to have a conbination of RAIN5 and RAIN1 depending on the VM I have setup between these. In the RAIN5 would have have 30 usable spindles? And then RAIN1 I'd have 20 usable spindles?
|
|
|
|
|
Supreme Being
Group: Moderators
Last Login: Yesterday @ 2:22 PM
Posts: 101,
Visits: 354
|
|
The 2 TB is how much each individual VSA can use as its internal storage.
So on your example servers you could have 1 VSA on each, as you say, and each should be given the max 2TB as a single VMDK.
That gives you a 3 nodes cluster with 6TB raw, 3TB usable with 2 way replication (what you called RAIN1)
You can create as many volumes of whatever size you want out of that = not limited to 2 TB. Create 1 3TB volume would work. With thin provisioning you can go ahead and create 10TB and add the nodes to support it later. Just trying to illustrate the use.
You mentioned RAIN5, but we don't support that in the VSA. SAN/iQ does "no replication" think RAIN0, "2 way replication" think RAIN1, "3 way replication" think RAIN1 + an extra copy, and even "4 way replication" think RAIN11. You'll probably just want to do 2 way replication on everything unless it is important enough to do 3 way.
You can't do 4 with only 3 nodes.
RAIN5 would perform at about half the speed of RAIN1 and would only yield you 16% more capacity in a 3 node configuration.
All volumes are spread across all spindles evenly so they'd all use every one of the 30 spindles.
Take this as a wild *** guess but you should see about 2000 to 3000 random IOPS out of that configuration. Bandwidth I'll not guess at because that is all up to your network, which if you get 10GB functioning correctly should be really good.
Adam C
Product Manager
LeftHand Networks
|
|
|
|
|
Forum Newbie
Group: Forum Members
Last Login: Thursday, July 31, 2008 1:19 PM
Posts: 3,
Visits: 15
|
|
So I'd just miss out on the extra 700 GB on each system?
I mean...could I create (2) VSA's or something with the same license so that the other 700GB would be usable?
IE: Is the 2TB a limit of VMWare or of VSA per se?
I know VMWare can only create a 2TB VMDK. But could I create another 700GB VMDK and still have it managed, used, RAIN'd, within VSA ?
RAIN just sounds cooler than RAID doesn't it?
|
|
|
|
|
Supreme Being
Group: Moderators
Last Login: Yesterday @ 2:22 PM
Posts: 101,
Visits: 354
|
|
You'd miss out on the extra 700GB unless you used 2 VSA per system, but you can not use the same key twice so that means 6 VSA licenses.
The 2 TB limit is imposed upon the VSA purposely by us by limiting the VSA to a single vmdk for data.
The main reason the VSA was limited to 2TB was that allowed for the majority of use cases we intended the VSA for but prevented situations like putting a single VSA in front of a 10TB SAN from some other vendor as an iSCSI gateway.
I've seen enough evidence of perfectly valid use cases that want bigger VSAs to convince me we should raise that limit. You case is a great example.
I can't say when we'll release a version of the VSA that allows for more than one virtual disk though.
Adam C
Product Manager
LeftHand Networks
|
|
|
|
|
Forum Newbie
Group: Forum Members
Last Login: Thursday, July 31, 2008 1:19 PM
Posts: 3,
Visits: 15
|
|
Adam...thank you so much for your help.
I agree the limit should be raised. A 10TB array really isn't much these days.
I know this is an unofficial forum. If I buy the Lefthand VSA's as I plan to do, and obviously get on maintenance....
What do you think the chance of that happening in say .... a year is? [upgrading of VSA to accomodate more setups like I'm trying to do]
50/50? 70/30? Just a guess....I know I can't hold ya' to it.
The thing I note is it appears the only folks really competing with you guys is SANMelody. And they don't have the RAIN ability like you do. BUT they do a lot of caching stuff so I'm sure the performance is there. (yes....it might be dangerous...but really???? nah....)
What would be ideal is if you guys would just go full boar on the VSA's. It would differentiate you in a crowded market. (even though it could be argued successfully that you are different already) I would say don't limit it. Keep the pricing the same. Get people addicted to it IE: VMWare. It's like meth to IT folks right now. And their pricing has risen for the enterprise stuff. Gone down for other stuff. Just my .02 cents.
(oh..and if you could make that happen within say... 30 days that'd be ideal?
|
|
|
|
|
Forum Newbie
Group: Forum Members
Last Login: Friday, August 01, 2008 12:59 PM
Posts: 1,
Visits: 4
|
|
We're considering the VSA for our DR site and branch offices, but the 2 TB limit is obviously quite a crusher when we want to store many snapshots from the 5 NSMs we have in our priamry datacenter. Running multiple VSAs on the same host sounds like reasonable solution (ignoring cost), except that really prevents failover from working properly when you lose a host (if it was running two VSAs, and you have two-way replication on volumes, then you're toast, right?)
Is it possible to set the data VMDK to be a Raw Device Mapping in VMware instead of a VMDK file? This would theoretically improve performance, and also allow for volumes larger than 2 TB. Or is the 2 TB limit hardcoded into the VSA (i.e. not based on the VMware limit for a VMDK file)?
Finally, I would offer the opinion that the VSA could commoditize iSCSI SAN storage in a way that would vault LeftHand past EqualLogic and NetApp in market share if some of the limitations are removed. Yeah, the per-unit margins might be lower. But would you rather have 50% of a 10B market, or 3% of a 40B market? Allowing customers to avoid rip-and-replace of their DAS gear would be an enormous selling point. I suggest you support more data in each VSA, and support more hypervisors (at least Xen and Hyper-V). You could be as important as VMware in the next-gen virtualized/clustered datacenter.
|
|
|
|
|
Supreme Being
Group: Moderators
Last Login: Yesterday @ 2:22 PM
Posts: 101,
Visits: 354
|
|
Thanks for the tips 
I did want to answer your particular question about this...
if it was running two VSAs, and you have two-way replication on volumes, then you're toast, right?
That is right, sort of, it depends on how you order the NSMs, oops, I mean VSAs.
With two way replication you only lose data or availabilty if you lose two consecutive nodes.
Say you have a 6 node cluster (1,2,3,4,5,6).
You can lose / reboot 1,3,5 at the same time or 2,4,6 and not lose data or availability. This can be leveraged by ordering the units correctly to get multiple VSA on the same ESX node and not lose availability if it reboots or dies a horrible death.
I'll give you one example that might help illustrate this rather complex point.
I helped build an 8 node VSA cluster once out of 5 physical servers. Why 8 nodes? Because 3 of the physical servers had twice the data capacity of the other 2.
So I had, Server 1= 1TB, Server 2= 1TB, Server 3= 2TB, Server 4= 2TB, Server 5=2TB
The customer did not want any space wasted. Easy answer was to put a VSA on each using up all space and create one cluster out of the first two and a different cluster out of the bigger 3. They did not want two clusters though, Sigh.
So instead, I put 8 VSA out (that did cost more) that had 1TB in each, Servers 3,4,5 had 2 VSA on them to consume the 2TB. The cluster order was.
Server 1 VSA
Server 3 VSA 1
Server 2 VSA
Server 4 VSA 1
Server 5 VSA 1
Server 3 VSA 2
Server 4 VSA 2
Server 5 VSA 2
Managers were run on Server 1 VSA, Server 2 VSA, Server 3 VSA 1, Server 4 VSA 1, and Server 5 VSA 1
That made one 8 node cluster that had all 8TB. Two way or higher replicated volumes, and quorum, survive any particular server going offline.
Now I bet you regret ever asking 
Adam C
Product Manager
LeftHand Networks
|
|
|
|