If Gigabit must be in your short-term future, the good news is that NetApp clusters do support it and it sure is fast! The bad news (you knew it was coming <g>) is that we don't support multiple MAC addresses per NIC with our Gigabit drivers today and consequently you'll need at least two Gigabit interfaces (one active, one passive) per filer in a cluster config. Again, no word on when that will change.

It won't ever change, because the drivers can't do what the hardware doesn't support. We've looked at support for the AceNIC 2, which I believe can handle multiple MAC addresses in addition to having much higher performance, but the engineer working on that has been focused on higher-priority projects.

-- Karl

+--- In a previous state of mind, kls@netapp.com (Karl Swartz) wrote: | | It won't ever change, because the drivers can't do what the hardware | doesn't support. We've looked at support for the AceNIC 2, which I | believe can handle multiple MAC addresses in addition to having much | higher performance, but the engineer working on that has been focused | on higher-priority projects.

I for one would rather have a single gigabit nic per filer as opposed to 2 per filer. In large clusters this gets real expensive quick. At $3k per NIC, seems like a better supported card is in order...

Sure. And our customers who can actually push an F760 hard enough to saturate a single AceNIC 1 would probably rather not have to buy two just to get the full bandwidth.

On the other hand, there probably are a lot of customers who would rather have failover *now* even if it only works with 100base-TX or FDDI (or requires extra GbE NICs). If we told a FDDI customer that we'd be able to give them failover today but we won't because we've not finished a driver for a GbE NIC that can handle multiple MAC addresses, they'd probably be a bit peeved.

Still other customers are happy with the current GbE (or don't care) and don't care about failover, but want other features. Presumably we've decided that it makes more business sense to give some of those wishes higher priority than failover *and* GbE *and* a single NIC.

So, your options at the moment are (a) use GbE but not have failover (b) use failover but not GbE (c) buy a second GbE NIC and have both failover and and GbE (d) wait until we have a GbE NIC and driver which will allow you to have failover with a single GbE

You'll be able to have your cake and eat it, too, but you'll just have to wait a bit.

Plus, what about filers with only single 64bit slots?

That would be the F630. On that filer, we only use the 64-bit slots (there is a second, but it must contain the NVRAM) in 32-bit mode because of some chipset problems. Still, that's the only available primary PCI slot, so the second card would have to go in a secondary slot. On the other hand, if you've taken over for a failed partner, you'll have degraded performance anyway.

-- Karl Swartz - Technical Marketing Engineer Network Appliance Work: kls@netapp.com http://www.netapp.com/ Home: kls@chicago.com http://www.chicago.com/~kls/

It is worth noting that this hardware bug is mentioned nowhere in your 'Technical Marketing' material. Thanks for saving the juicy ones.

Nor do we say anywhere that we make use of a 64-bit slot as such. I know we also didn't publish the LADDIS run on the clock-doubled F330 which showed essentially zero performance gain over the stock model, nor do we explicitly say that we don't use the hardware's ability to fault in a paged-out hunk of memory even though we all know that even a 486 could do that. I'm not sure what the point would be of trying to mention every little hardware nugget that we don't choose to take advantage of, so long as we don't misrepresent the product as gaining from such features.

-- Karl Swartz - Technical Marketing Engineer Network Appliance Work: kls@netapp.com http://www.netapp.com/ Home: kls@chicago.com http://www.chicago.com/~kls/

+--- In a previous state of mind, Paul Norman pnorman@netapp.com wrote: | | Err, well in order to have a Cluster, you need to have a Cluster Interconnect | card and yes, you guessed it, it has to go into the unbridged slot - so there | really aren't any available for high speed NICs.

So, are you saying that with GbE and a f630, you cannot do failover?

Failover to another interface would be a good feature to have (esp that all the 7xx have 10/100 built in). It is probably not simple to setup, but it would at least make it so I would not have to get dual GbE nics. Then again, if I am dropping an order of magnitude in speed, it may not be useable anyway.

Alexei

+--- In a previous state of mind, kls@netapp.com (Karl Swartz) wrote: | | FDDI (or requires extra GbE NICs). If we told a FDDI customer that | we'd be able to give them failover today but we won't because we've | not finished a driver for a GbE NIC that can handle multiple MAC | addresses, they'd probably be a bit peeved.

Well, I am a FDDI customer and I would rather have the GbE driver. I am migrating from FDDI to GbE now. So, for clustering, you have just increased my costs by 50%. How? Well:

sofware for filers & interconnect cards: $30k

additional gigabit cards: $6k additional ports on GbE switch: $8k additional cabling $1k

So, do I want you to complete the drivers for a single card? Damn right.

| (a) use GbE but not have failover | (b) use failover but not GbE

I wish I had known about the delay of GbE and failover *before* I bought it.

| (c) buy a second GbE NIC and have both failover and and GbE

See the above financials fo the reason not to do this.

| (d) wait until we have a GbE NIC and driver which will allow you | to have failover with a single GbE

When will this be?

| You'll be able to have your cake and eat it, too, but you'll just | have to wait a bit.

As long as the windows folks had to wait for acls? I certainly hope not..

| slot. On the other hand, if you've taken over for a failed partner, | you'll have degraded performance anyway.

It depends. If my load is spread such that neither filer is foing more than 50%, then I should not be in degraded mode, right?

Alexei