Question About V-Max
Was just thinking through some of the newer storage tech today while driving to and from a customer. It is widely expected that HDS is going to announce the availability of their clustering of the USP-V line of boxes tomorrow. We know that the technology has been built into the microcode and hardware since the release of the box so I don’t think it’s going to be a huge surprise.
So I got to thinking about the V-Max and how it’s cache a disk is managed. I haven’t read too many of the manuals yet, so I thought I’d ask here since I know the EMC guys read this blog.
- What sort of traffic is traversing the RapidIO network today? Is it just management traffic or is IO actually being sent from one engine to another?
- If IO is sent to one engine but the LUN is managed by another engine then how is cache coherency managed? Is the cache one large pool or does each engine have it’s own cache?
- Are the LUNs addressable in an active/active fashion as the DMX-4 and USP-V are handled today? e.g. Can an IO be sent to multiple engines at the same time in a true active/active fashion? We’re not talking about Clariion "active/active" here.
Here is the way I kind of see it. Each engine manages it’s own set of storage and forward IO to the other engines as needed. Is that right or wrong? I’m looking for an education here, so please help me out. To me it seems like since each engine has it’s own FE and BE ports that the BE ports must have a specific set of disks attached to them. Right?
May 27th, 2009 at 4:28 am
I can answer two of three questions, but I am not 100% sure about the first question, so I will leave that for another person to answer completely.
Depends on the configuration. I you are using one of the engines as an R21 device (diskless synchronous mirror that can forward the data to an asynchronous end point (R22 device), it is only natural that all of the IO get’s forwarded. In another setup I don’t think all of the data is forwarded to all of the engines. I couldn’t see how that would be a performing solution.
The cache is actually one large pool.
Ah, nice reference to ALUA mode. As far as my information goes, and from what I have seen on the boxes here, it is actually a true active/active setup.
As stated, depending on the setup, you don’t need to have disks attached to each BE (at least not usable, I am pretty sure you would need some for vaulting purposes). Basically when you add an engine, you just increase the maximum number of ports and the performance of the whole thing. If you have more ports, you might as well use them for performance, right?
Now, I hope that I’m not telling a complete load of ****, but I’m pretty sure that someone will correct me if I’m wrong.
May 27th, 2009 at 4:49 am
By the way, I’m using FF 3.0.10 and the markup box displays, but the markup I created and the smileys I entered seem to have vanished?
May 27th, 2009 at 6:39 am
Snig, your diagram is correct – each engine rules own set of DAEs with own set of drives. And it seems to be active/active (meaning you can send I/O to the LUN via any engine). I’m not sure though about intra-engine mechanics – whether it’s active/active or ALUA.
On the cache side: cache is global in the sense that the slot can be allocated on any engine regardless of which engine recieved I/O or manages disks. To facilitate this, V-max uses store and forward buffers in cache which seems to be the only area accessible to FE and BE ports. So when I/O comes, it’s stored into S&F buffer, then copied to relevant engine’s buffer then put into general cache. So, before I/O commits you always have several copies of the block in memory, possibly on several engines. As a result, RIO links should be pretty busy with real I/O, not just management traffic, and the bigger V-max is – the busier switches should be.
That’s about what I was able to get from local EMC guys.
May 27th, 2009 at 6:47 am
>Basically when you add an engine, you just increase the maximum number of ports and the performance of the whole thing.
Hrm… DAEs have two redundant loops. You can’t connect single DAE to more than one engine. So you have to add the engine if you need more capacity than your current amount can own.