Is your ISCSI fast enough?

July 30, 2014

iSCSI is a great technology. It gives you the ability to create SANs very cheaply and easily, without having to become a guru in fibre channel or put yourself into deep debt buying all the fibre equipment. By using easily available networking equipment you can add storage to existing boxes, even if you want to go crazy and do shared access, clustering or other high-end features.

A lot of vendors provide basically free iSCSI targets (there’s even one in Windows Server these days), and almost every OS has a free software initiator to connect to those targets. Yes, we can bang on about whether software or hardware initiators are better, but software are free and work so well that most iSCSI hardware initiator vendors have stopped bothering.

For the uninitiated, a target is the machine with the storage, and an initiator is the machine connecting across the network to that storage – simple (if you say it quickly but in reality it’s not that hard these days).

So now that I’ve oversimplified iSCSI, let’s look at it’s performance. It’s pretty good for most things, but in my experience suffers in the random read/write area – small read and writes carry quite a bit of network overhead etc.

So what can be done about that?

Hmmm … spend millions with a big-name vendor on a super-duper iSCSI target that you don’t need, or … cache the random reads and writes at the initiator end, so that those small, frequently used blocks don’t actually travel back and forth across the iSCSI network.

That sounds very attractive, and in fact it is very easy to achieve. Adaptec has a product called “maxCache Plus” – which is caching and tiering software, and is built into all 8 Series “Q” controllers (eg 8885Q). Now prior to this version of the software that works with these controllers, caching only worked on devices attached to the controller, which an iSCSI target definitely isn’t. However now with “maxCache Plus”, you can cache or tier any storage in your server, including iSCSI targets that are in fact disk drives sitting somewhere out on the network, but just appear to be drives inside the server.

So …

Plug in an 8Q controller, connect 1 or more SSDs (you don’t even need to make them RAID – you can use them as single raw disks), then add caching to an exiting logical device (disk) – using what we call “CachedLD” (Cached Logical Device).

That logical device can be an iSCSI target, an on-board RAID array off the motherboard, or even a RAID array from a competing RAID card vendor … and did I mention that you don’t need to reconfigure the existing data on the server, or make any changes to the server configuration? That sounds too good to be true, but in fact it works seamlessly.

So you could in fact make your own iSCSI target using cheap hardware and pretty much free software to create a very cheap network storage solution – then accelerate the random reads and writes to that locally in your server, rather than spending a fortune in the iSCSI arena. Hopefully you’ll use an Adaptec RAID card in that iSCSI box, but more than likely you’ll use free Linux software RAID – what the heck, just do it :-)

By the way … what happens if the SSD fails (in the case of using a single SSD) – absolutely nothing to your data – you just go back to original performance before installing caching – CachedLD cannot hurt your data.

Now that’s something to think about.

Ciao
Neil

 

facebooktwitterlinkedinmail

Playing games beats working … any day …

July 16, 2014

Was recently asked a question about SCSI-Express by a customer … eg when are we going to have controller out to handle such devices etc. So I flicked back a question … what devices?

He quick threw back some links of Asus and Gigabyte motherboards that have SCSI-Express connectors on them – and yes of course they are gaming boards (http://www.gigabyte.co.nz/products/product-page.aspx?pid=4959#ov).

I dug up an old link from the STA (http://www.scsita.org/library/documents/STA_TechShowcase_DataSheet_AdvanPCIe-print.pdf) and all you’ll find all over the wording is “enterprise”. If it had been written today, it would have the word “datacenter” stuffed in there as well because that’s the new marketing buzzword (especially “Hyperscale Datacenters”), but I guess that document must be more than 10 minutes old which makes it old hat in this world’s marketing analysis.

However, it made me think … who is going to use SCSI-Express and/or NVMe devices? Are they going to be so much cheaper than the current crop of straight PCIe NVRAM cards (eg LSI Nitro or FusionIO) that everyone will run out and buy one? When was the last time that people put FusionIO cards in gaming machines?

Maybe I’m missing the point. Maybe the gamers of the world do indeed spend that much money on storage. Somehow I thought they spent all the money on processors, video cards, memory and fish-tank cooling systems … not PCIe SSD.

Now I’m not saying what we are doing in this department because heaven forbid, that would get me in hot water, so you can just wait for the marketing team to talk about “real world” stuff, but seriously, am I getting too old?

I’m confused … is there a “gaming enterprise” segment of this market that I hadn’t heard about.

Ciao
Neil

facebooktwitterlinkedinmail

Content delivery …

July 16, 2014

We have a constant battle in the halls of PMC … what exactly should the blog look like? While some of that discussion is based around the look and feel, most of it is based around the content.

Should it be technical? … a real propeller-head’s delight?
Should it be generic in nature (in my opinion boring as …)
Should it be perceptive and propose the future (I can hear legal fainting from here)
OR
Should it be a combination of all these things?

I think it should have lots more pictures of Ducati motorbikes (my pride and joy), but the marketing team don’t really agree on this. So here’s your chance … tell me what it is you want (that’s dangerous I know) … and if it’s physically possible I’ll give it a go.

Meanwhile I’m going to sit back and look at the bike because it’s way too cold to ride the thing at the moment :-)

Ciao
Neil

facebooktwitterlinkedinmail

So who is using whitebox servers?

July 8, 2014

We have this discussion all the time at Adaptec … who is using whitebox servers?

Focus number 1 these days is datacenters – and yes a large portion of them use whitebox. Once they get to the numbers that those guys use it makes sense to have them built for you by an ODM in Taiwan etc – the brand name players get a bit expensive in those numbers.

But at the other end of the spectrum, I recently came across a group of organisations who also promote and use whitebox servers and I must say I was a little surprised.

Managed Service providers. A lot of these guys provide total solutions to small to medium business (and some pretty large business as well) … they do the telecoms, the networking, the internet and these days are pushing to provide the hardware for servers and workstations either installed in the customer site with remote management, or servers sitting in their own small/medium-sized datacenters.

So why do they want whitebox? Mainly because they don’t want to have to call the brand name hardware vendor every time they think they have a hardware problem, and they don’t want that brand name player coming to the customer and promoting their own services (lots of competition in this market).

So it’s better to use whitebox (generally cheaper) – they have control over the build of the box to suit the customer, and they have control over all components from software to hardware, and can provide the service they want to a customer without bringing in a third-party brand-name vendor to replace a hard drive etc.

Makes sense to me.

So who else is using whitebox out there and what are you using them for?

Ciao
Neil

facebooktwitterlinkedinmail