IBM may have come late to the converged infrastructure party but it’s leveraged its decades of experience in virtualizing workloads to create the first in an ongoing series of integrated offerings namely PureSystems. Here’s a closer look at what this entails. By Prashant L Rao
When Cisco got together with EMC and VMware to launch Vblock, arguably the first converged infrastructure offering to hit the market back in 2009, it set off a chain of events. The same year saw HP launch its CloudSystem Matrix, a full-fledged rapid deployment private Cloud offering. The server majors that were conspicuously absent from this party were IBM and Dell. In Big Blue’s case, it turns out that the company began working on what it calls Expert Integrated Systems and, in true IBM fashion, it went about the job diligently starting with its researchers and moving on from there.
It’s interesting to note that PureSystems was a reaction to the growing commoditization in the x86 server market. Kashish Karnick, Product Manager, BladeCenter Systems, India/South Asia, commented, “IBM created PureSystems with the intention of differentiating itself from the pack in a crowded x86 server market where a server has become a commodity.”
While virtualization does confer flexibility and results in higher utilization of the underlying hardware, it also brings with it performance overheads and complexity.
Now you see it, now you don’t
“Network switches can’t see VMs and using virtual switching adds a performance overhead. Today, you can’t handle more than eight VMs at a time. Then there’s the fact that when you move VMs, the network settings don’t get copied. The ideal situation is that you want things to be as seamless as they used to be,” said Karnick outlining some of the issues with contemporary virtualization solutions.
When it came to defining the value proposition for PureSystems, IBM went about doing one thing which it has decades of experience in, viz, to simplify the chores that an IT team has to put up with after deploying virtualization. “Virtualization has made a sysadmin’s life more complicated. They spend more time configuring things than before,” said Karnick.
IBM’s researchers went about the task of coming up with an integrated system by devising workflows for virtualization for tasks such as upgrading firmware, load balancing, application deployment etc. A glance at the company’s hardware/software stack revealed a lack of networking know-how. This was the trigger for the acquisition of BNT in 2010.
“Once we got that, we had the entire stack with middleware, database et al,” said Karnick.
Source code
Another area where IBM claims that its stack scores over the competition is in the fact that, since it owns all the components of the stack, it has source code access to it. “Today all the integrated stacks, other than Oracle’s, are built around API-based communication,” said Karnick.
While PureSystems is the first family of integrated systems from IBM, it’s not going to be the last. Others are on the way. Within PureSystems, there’s the infrastructure system and the application system. The vendor is positioning the infrastructure system, PureFlex, as a panacea fro management issues. For application deployment, it offers PureApplication.
Within the PureFlex there are the compute, network, storage (StorWize v7000) components alongside Brocade and other storage switches. For the Cloud layer there’s an enterprise Tivoli Cloud that’s in PureApplication. There’s also an entry-level Cloud called SmartCloud Entry in PureFlex.
Another area that IBM focused on was with regard to having a single console for managing the entire system. With the goal of addressing scalability, which is a problem with virtualized systems today, it designed the central consoles in a particular manner. “We decided that the central consoles would only dictate policies so that the work gets delegated closer to where it’s being done. The biggest challenge in IT today is scalability. Beyond 30-40 VMs, performance goes for a toss,” explained Karnick.
With PureFlex the Central Console shows you the VMs, network and storage. For database VMs, heartbeat is a network property, storage IOPS is a storage property, VLANs are a network property—you can capture all of this and cut and paste the information.
Virtualization is quite popular in sectors such as IT/ITES where companies have lots of projects and they need to constantly provision/deprovision resources in a jiffy. In other verticals, adoption has been piecemeal with select instances of VDI etc driving the uptake of this technology. Having said that, there’s a reason as to why both IBM and HP are taking this slice of the market so seriously. It’s the simple fact that the number of virtualization deals may be low but the deal size is high and the impact on the customer is humongous. “That’s why it’s a high focus area,” quipped Karnick.
On the storage front, there’s the StorWize V7000, which is a storage box plus a SVC appliance. An enterprise can buy just the controller and use other vendors’ storage boxes. The licensing is per enclosure and not per TB. That’s if you go in for SVC. If you don’t and prefer to use Brocade or Qlogic switches, you can connect to only one storage box. In this way, PureSystems is open on the SAN front.
“The customer’s DR becomes cheaper. You can use FCIP routers to replicate across sites with minimal bandwidth utilization,” said Karnick.
In terms of compute, IBM has always had more platforms than any other vendor and, naturally, it’s leveraged that to offer x86 and Power—with both Linux and AIX on the latter.
The networking bit is quite interesting. As mentioned earlier, conventional network switches can’t see into the VM. This puts the burden of dealing with virtual network traffic on the hypervisor.
Qbg or Qbh: a tale of two protocols
IBM has adopted a protocol called IEEE 802.1Qbg that’s also known as Edge Virtual Bridging, which allows the switch to see virtual network traffic. It claims to be the first company to launch switches that support this protocol. Other vendors supporting this protocol include HP, Dell , Juniper, Brocade and Qlogic. The notable exception here is Cisco which supports a similar protocol called 802.1Qbh or Bridge Port Extension. Cisco’s popular Nexus 7000, 5000 and 1000V all support Qbh.
“If you have a Cisco Nexus switch and an IBM Qbg switch, they can talk to each other on Ethernet protocols, but you can’t monitor the VM network traffic,” commented Karnick.
If a company goes in for a 100% Cisco stack with UCS servers and Nexus switches, everything’s fine. “The problem is that you are moving all your VM traffic to the core. Your normal Ethernet traffic wouldn’t exceed 400-500 Mbps. Here it goes to 4-5 Gbps. So you end up buying more switches. If a VM was 100-200 Mbps and you have eight VMs, that’s 1.6 Gbps. With Qbh, the philosophy is that everything is a Nexus stack and you get control but you have to move the VM traffic up. There’s user and IT traffic on a network. VM traffic is IT traffic and you want to keep it as close to the host as possible. That’s the main difference between the Cisco architecture and everybody else. If you have a Qbg switches from IBM, Juniper or HP, you can move your VMs across the data center with no issues,” said Karnick.
On the hypervisor front, IBM supports KVM, PowerVM with source code integration. With VMware, Citrix and Hyper-V, it offers API-based integration.
“If I want to automate a firmware upgrade, a server with 20 VMs, you can only move eight at a time with VMware. You have to figure out which one to move first, where to move it etc. We have a 10 GbE port that can be spliced into four NICs. HP also has that technology. We can vary the bandwidth so that you don’t have to worry about VLANs, QoS or ACLs. It’s all automated. The VMs are automatically moved, the firmware gets upgraded, PureFlex checks if it’s green and, if it is, then the VMs are moved back,” he explained.
“If you want to do energy management, you have to manage the server components. VMware can’t do that. In our system, you can specify that if the energy consumption goes above a particular level for a sustained period of time, you have to move the VMs off,” he added.
The bottom line, as per IBM, is that PureSystems adds up to a simplified experience for a sysadmin. “We have a feature called Quick Find that lets you search by VM name, IP address, part number etc,” said Karnick.
Today, the deal sizes for PureSystems range from $150,000-300,000. “It’s only when you grow beyond 70-80 VMs that it makes sense to do something like this. So far, 50% of the PureFlex deals have gone with IBM storage,” said Karnick.
IBM has deployed VDI, deployments that did not involve virtualization, virtualization with SAP, deployments on Citrix and Microsoft Exchange on PureSystems. It’s targeting SMBs too since the larger ones are deploying 70-80 servers. “As business users get comfortable with IT, the app layer grows and things start to become complicated. That’s where virtualization can help. Wherever there’s low utilization and you have multiple apps, you should virtualize,” advised Karnick.
Converged infrastructure is a category where competition is intense. The developments resulting from IBM throwing its hat into the ring will be far reaching. This is shaping up to be another battle—similar to what happened in the case of network storage a decade back—where the server and storage vendors face off. You have both EMC and NetApp allied with Cisco & VMware on the one side. On the other you have the full stack players like IBM, HP and Oracle. It will be interesting to see who wins this one.
