What Is Data Caching? The Basic Concept
To understand I/O Express, you must first understand the concept of data caching and the enormous performance benefits it can provide. So, before we ask "What is I/O Express?" let's first answer the question, "What is data caching?"
Data caching is a concept that makes use of the assumption that, of the great amount of information that it is possible to process, only a very small amount of that information represents the vast majority of what is actually processed.
Caching utilises a relatively small amount of high-speed memory to allow quick access to the information most often used. Without caching, the information would have to come from relatively slow devices - disks.
Caching effectively keeps a small amount of the most actively used information available so that it can be accessed far more rapidly. This means that you get a lot more processing done. Reading from cache memory typically takes between 0.5 and 2 milliseconds including operating system overhead. Reading from disk may typically take 8 to 20 milliseconds. Disk access speeds vary, but it is always much, much faster to read from the cache.
 |
Types of Data Caching
OpenVMS has utilised a number of types of caching for many years - the concept itself is not new. But without exploring all these different types of caches here, suffice it to say that they all had one thing in common: They utilised a fixed portion of memory, and generally cached the same data on a consistent basis.
So, what is I/O Express 4.3, and how is it a vast improvement on these other methods? Read on!
What Is I/O Express?
In 1989, Executive Software released I/O Express Automatic Data Caching Software, and it introduced a whole new concept: intelligent data caching! I/O Express went far beyond the utilisation of a fixed portion of memory to cache pre-determined data.
Determination of Data to Cache First
It was decided that I/O Express would do better to cache data blocks, as opposed to files. In many instances, especially with large files, only small portions of those files would be frequently accessed. Therefore, I/O Express was designed to cache the most frequently accessed data blocks instead of files.
Second, a unique algorithm was designed to determine which data blocks were the most frequently accessed. Leaving this determination up to a human invited all sorts of problems, in that manually determining which data is most "frequently accessed" could be an enormous chore, especially on large systems.
Memory Management
Next, it was realised that allocation of a fixed portion of memory for caching would not be efficient. If other processes needed the memory so assigned, that memory would have to be manually de-allocated, thus rendering the product difficult to use.
I/O Express was designed so that only idle memory would be utilised for caching. And since the amount of idle memory fluctuates, so does I/O Express' use of memory. If memory is needed for idle processes, I/O Express backs off and returns memory to the system. When that memory becomes idle again, or if more memory becomes available, I/O Express utilises it.
How Well Does it Work?
The gains available from I/O Express can be measured in numerous ways.
The most noticeable gain, of course, is performance. We have many, many success stories about happily raving users, batch job times cut to a fraction of the "norm," and performance more than doubling. This certainly makes I/O Express well worth the price!
In terms of a technical measure, I/O Express is best measured by "hit rates" - the rate of times I/Os are successfully fulfilled from cache.
This measure is used by I/O Express to measure its success, and so a system manager can monitor how much gain is being obtained.
Additionally, the use of I/O Express shortens disk I/O queues, because the most frequently accessed data is now in cache and being accessed from there. So, not only does I/O Express speed up overall I/O performance, disk I/O performance is sped up as well.
How much of a performance increase is possible? Well, if your cache gets a 50% hit rate and reduces the average I/O queue by 1, you can get over a hundred-fold improvement in disk access and I/O, and hit rates of 50% or more are not unusual. Some applications have gotten over a 90% hit rate and further reductions in I/O queue length. We have seen some I/O-intensive applications become so efficient that they become CPU-intensive applications.
|
Cluster Performance
The whole point of a cluster is, of course, shared computing resources. With a large database, for example, the data can be stored on a disk farm and accessed by hundreds of users through different cluster nodes, greatly increasing the efficiency of existing equipment.
I/O Express was created with clusters in mind from the outset. In many cases, within a cluster a user is logged into a certain node, and the data must come from a disk or disks attached to another node or nodes. Many users are accessing the same data on the same disks from various nodes in a cluster.
In a clustered environment, I/O Express caches the most frequently accessed data in the memory of the node or nodes that need it. Not only does this eliminate the extra cluster communications needed to access the data remotely, but also the need to constantly retrieve the data from the disks. I/O Express can be a tremendous boost to cluster performance, as well as a great cost-saver.
Safety
As with any software product, safety is a prime concern. I/O Express never puts your data at risk. Data read into the cache is, of course, still resident on the disk. Data written to the cache is written through back to the disk immediately, and the cache is updated when the data is again requested. To keep the cache current, an invalidate message is sent to all member nodes so that the previous version of the block just written will not be read from cache; instead it will again be retrieved from the disk when needed.
Low Overhead
I/O Express accomplishes all of this at virtually no cost in terms of system resources! The memory used by I/O Express would not otherwise be used, and when workload changes dictate, I/O Express returns memory to the operating system. CPU time to satisfy an I/O request from memory-resident cache of I/O Express is about the same as it would be without I/O Express - on some configurations, even better!
Who Can Benefit from Data Caching?
To benefit from I/O Express there must be some memory available some of the time for caching. However, I/O Express is so efficient that it can utilise memory when it is available and give it back when it is needed elsewhere by he system.
Applications most likely to benefit from data caching include:
- Database applications including those written in Oracle, INGRES, Rdb, and other relational databases.
- ISAM file based applications.
- CAD/CAM applications.
- On Line Transaction Processing (OLTP) applications.
- All-In-1 sites.
- Many financial and commercial applications, and much, much more.
Application Software Compatibility
I/O Express 4.3 is compatible with all application software and all types of Alpha and VAX configurations. It can be run on any and every OpenVMS system safely.
Since caching is done at the I/O request level, I/O Express is completely independent of files and applications and fully compatible with RMS and other record management software applications.
|
Automatic and Simple, Yet As Flexible as You Need
I/O Express is a "Set It and Forget It"® product. Once installed, you can start I/O Express with one command, and it immediately begins caching. You'll start seeing gains right away.
In addition to ease of startup, you're going to want the ability to control caching as needed. I/O Express gives you complete control over which disks on the system are being cached. While all mounted disks are cached by default, you have the option of disabling caching on any disk at any time. You can also enable caching on any disk at any time. Both of these functions can be done "on the fly," while I/O Express is up and running.
|
Buy it Today!
I/O Express 4.3 can give you a dramatic performance increase at the fraction of the cost of a hardware upgrade. Call your sales rep at +44 (0)1342 327477 and order it today!
Or order now from the Online Store.
Can Alpha Really Benefit From I/O Express?
Over the years, Executive Software has conducted many surveys of the OpenVMS marketplace. One of the questions we've asked on many occasions has been, "How do you handle slow system performance?" The majority answer has almost always been, "Faster CPUs" or "better hardware" or "more hardware."
In 1992, it seemed as if Digital Equipment Corporation was listening to these answers as it released the fastest computer chip ever: The Alpha. At 150MHz (at that time; it has since exceeded 500MHz) it delivered the best price/performance of any CPU around.
Now, with this super-fast chip one could ask, has the problem of slow performance been solved?
Before we answer that, let's take a look at one of the common causes of system slows.
|
A Scenario
Let's invent a hypothetical site - the XYZ Company. We'll say XYZ's System Manager has been going out of her mind from user complaints - the system is too slow! Overnight batch jobs are taking so long they're cutting right into morning production! It takes two minutes to get a prompt!
Hours and hours of system tuning and tweaking (things such as adjusting system parameters and load balancing between CPUs and disks) has only provided marginal improvement. Finally, our System Manager throws up her hands and says, "That's it! Got to add another CPU!"
At this juncture the System Manager could go through the evaluation and cost-justification and requisition-writing needed to obtain the new machine.
But wait. Since this is all hypothetical anyway, let's say that before she went to all that trouble, she discovered an information packet on her desk she never got time to examine - a packet of information on I/O Express Automatic Data Caching software. With the information, she finds a free demonstration copy of the software.
She first reads the product information and runs across something interesting: Disk I/O bottlenecks are a common cause of slow systems.
This of course she already knew. But what she didn't know was that a new CPU would not truly handle the problem - and best of all, there might be a far less expensive way to solve her slow system. A software solution.
She loads the evaluation copy of I/O Express. Within a few days, batch job times have been cut in half or better, access time has improved markedly, and users are calling in with compliments!
That's one costly hardware purchase delayed. And, with I/O Express, there have been many of these. We have the success stories to prove it.
|
Data Caching on Alpha
So, in our hypothetical site, what would have happened had our System Manager gone ahead and bought that new CPU, without first trying I/O Express?
Well, first, the gains available from data caching would have been bypassed. And second, while performance would of course have been improved with the new CPU, the gains available would have been dampened due to the fact that the original problem - disk I/O bottlenecks - still existed. Eventually, it would become a problem all over again as the new machine became more and more utilised.
The underlying problem is this: No matter how fast the CPU, your disks are still operating at the same or about the same speed as they always have. While CPU performance is roughly doubling every year, disk drive performance is only doubling every decade. As a result, today's CPUs can process much more data in any given length of time than today's disks can provide in the same length of time.
So, out of our hypothetical site and back into the real world: Now that all these Alphas are out there running, have slow systems due to I/O bottlenecks been solved? Clearly, based on the above information, they have not. In fact, due to the widening difference between disks and CPUs, they have been intensified.
I/O Express' intelligent, automatic data caching is the answer. It automatically identifies your most frequently accessed data blocks, and places them in a cache in memory. Data retrieval from memory is much faster than from disk - upwards of 10 times faster. Retrieval from a disk takes, on average, between 6 and 20 milliseconds. Compare that to retrieval from memory, which takes from 0.5 to 2 milliseconds. You can see the potential for dramatic improvement, especially in large sites with commonly accessed data.
|
Put I/O Express to work on your Alphas today!
Comparing Data Caching Products: How do you decide?
Many OpenVMS system managers require that intelligent data caching be part of their system configuration these days. Not surprising! With CPUs as fast as they are, and disks as slow as they are, caching is vital in today's demanding enterprise computing environment.
How do you decide on which caching product to run? They come with a variety of Funktionen and are available at a variety of prices (one is even free).
Before we get into comparisons, you should remember that there's no comparison better than running different caching products on your own system and comparing them yourself. To aid you with this, we offer free 30-day trial copies of I/O Express, an offer you should take advantage of prior to making any decision. Remember, we wrote the book on intelligent caching; we were the first, and are still the best. At least give the best a try before you decide.
|
PERFORMANCE
With any caching product, performance is obviously the object. This performance is obtained by caching the most frequently accessed blocks so that they are available from memory - access which is well over 10 times faster than from disk.
Any caching product can claim to give you the best performance, but again, the most basic test of a caching product's performance is running it yourself, right on your own system. There's nothing like hard numbers to tell the tale. Do yourself the favour of testing a caching product on your system before making your decision.
SAFETY
Of primary concern with any software product is, will data be lost or corrupted? With I/O Express, no chances are taken with your data. When data is cached, the cache, of course, is a copy of the data still resident on a disk. When writes occur, the write goes straight to the disk, and the cache is updated from the disk when needed.
Make sure any caching product you select at least meets this level of safety.
EFFICIENCY
While evaluating for system speed, ensure you also take into account the resources used by the caching product. Some caching products increase performance, but at what price to your other high-priority jobs? A caching product, to do its job efficiently, must do so with the least negative impact on system performance. So, ensure you examine both points - performance increase and negative impact on system resources - before deciding on a caching product. When examining a product for efficiency, you should, at minimum, examine the following points.
|
MEMORY USE
Early caching products were extremely inefficient with memory use, using only a fixed amount of memory which needed to be enabled or disabled manually. I/O Express was the first product to utilise memory only as it became available, and to back off when memory was needed for higher-priority processes.
Ensure any caching product you select is efficient with memory. With I/O Express, you will have the most efficient use of memory on the market today. I/O Express utilises only available memory, which means you won't be short of memory when required by higher-priority jobs, and you'll obtain even more of a performance increase when there is more memory available for I/O Express to use.
Additionally, make sure any caching product you select doesn't require a relatively large amount of memory to do its job at all. I/O Express can begin caching even when memory usage is quite high, even as high as 100%.
|
Caching Across Clusters
Today's computing environments demand that cluster-wide caching is available. What's the point of having a caching product that will only cache on local nodes? This negates caching benefits for any user or process on a cluster accessing a file or part of a file remotely.
I/O Express is capable of caching any disk on any cluster, for any user connected to any node in that cluster. Having a file open and cached on more than one node of a cluster is no problem - when a write occurs, the write first goes to the disk, and an invalidate message is sent out to any nodes caching that same file. When the data is again accessed, it will be accessed from the disk and newly read into the cache.
There are some caching products that will not cache data from any file that is open on two more nodes, if that file is opened to write new data. Since opening a file to write new data is very common with applications such as databases, this makes such a caching product's usefulness in a cluster very limited. It will only cache a file in a cluster if that file is never written to. How often does that happen?
Types of Blocks Cached
Most data caching products cache data blocks to maximise performance. But you should take care to examine what type of blocks cached - there's more than one kind of block, and the type of block cached can make quite a difference. A virtual block is a block numbered in sequence according to its relative position with a file instead of by its physical location on a disk. A logical block is a block numbered and tracked by its relative position on the disk. The Logical Block Numbering system is the system used by OpenVMS to locate data blocks on a disk.
Some caching products cache file virtual blocks. In other words, they only "see" data as it relates to individual files, as opposed to the entire disk. The difference in resource consumption between these two methods is quite drastic. In order to cache on a virtual block level, the cacher has to be much more resource-intensive and a lot more intrusive on the file system itself. This is due to its use of the OpenVMS Distributed Lock Manager for its synchronisation between caches in different nodes.
I/O Express only caches logical blocks. In doing so, it is not restricted to file level or Distributed Lock Manager. It does its synchronisation through SCS (System Communication System), which is far more resource-efficient than Distributed Lock Manager.
Types of blocks cached is an important point to examine when deciding on a caching product.
|
EASE-OF-USE
"Set It and Forget It"
System management is a terrifically busy job. A system manager, constantly badgered to tweak system performance, add new hardware, test and add new applications, and so on, has no time to worry over a product that should be taking care of itself.
I/O Express, of course, is a "Set It and Forget It" product. Once installed, I/O Express can be started up with one command, and will efficiently cache from there on out.
How much "configuring" and "tweaking" does a caching product need to get it up and running? This is a valid question when evaluating any caching product.
|
Flexibility
In addition to ease of start-up, you're going to want the ability to control caching as needed. I/O Express gives you complete control over which disks on the system are being cached. While all mounted disks are cached by default, you have the option of disabling caching on any disk at any time. You can also enable caching on any disk at any time. Both of these functions can be done "on the fly," while I/O Express is up and running.
Being able to modify caching while the caching product is up and running saves valuable time, especially compared with a product that, once it is adjusted, can only effect those adjustments when the system is rebooted.
Other caching products offer differing levels of flexibility with cache control, ranging from pretty good to fairly ridiculous. One caching product offers no cache control at all, save having the product turned "on" or "off." This causes a situation where data that is seldom (if ever) needed is wasting system memory that would be better utilised caching frequently accessed data. And when it is "off," it is exactly that - nothing gets cached.
If you, like most system managers, are interested in being able to tailor caching to fit your system's needs, make sure the caching product you select is as flexible as you'll need.
|
SUMMARY
Before you select a caching product, check for these points to ensure you're getting the most for your money:
PERFORMANCE
Make sure your caching product actually increases your performance. This is best done by testing the product on your system.
SAFETY
A primary requirement of any caching product is that it will not lose or corrupt data.
EFFICIENCY
Make sure you check that a caching product is efficient in, Memory Use, Caching across Clusters and Types of Blocks Cached.
|
EASE-OF-USE
Make your the caching product you select won't take hours to set up and constant babysitting to run. On the other hand, make sure that a caching product contains the flexibility you need to adapt it to your particular environment.
I/O Express 4.3, of course, meets all of these qualifications! We invite you to find out just how well I/O Express will increase your performance and meet your needs. Call your sales rep right away at 1-800-829-4357, and order I/O Express today! If you want to try before you buy, take advantage of our 30-day free trial of I/O Express.
Cost Justification: When Is an I/O Express Purchase Worth It?
I/O Express has a unique position in the computer industry. It's not an application - it is an enhancement to the system itself. Therefore, the rules applied to its purchase can vary greatly from the purchases of other software products.
When is it "worth it" to buy I/O EXPRESS?
- When you've already seen a benefit. This would, of course, be the number one reason. You've run I/O Express, and you've seen your batch job times cut in half or better, or you've seen overall system performance improve substantially. You may have even had users comment on the difference.
- When you know you have a problem the product will solve. This reason also makes the purchase very "worth it." You've run our I/O Wait Monitor Utility and seen how bad your I/O bottlenecks are. Or, you've tried (demo'd) I/O Express on your system and seen a positive result. You're confident of the purchase because you know it will solve a problem you've proven you have.
- When you want to get the most out of your existing hardware. These days, budgets are tight, and it can be very difficult to cost-justify a hardware purchase. Yet, management and users are pushing to increase performance. It can wedge a System Manager between a rock and a hard place. Eliminating disk I/O bottlenecks is a very basic step in maintaining the native performance of a VAX or Alpha. Due to the difference in speed between the CPU and the disk, they can bring system performance to a crawl. Using I/O Express, your frequently accessed data is placed in memory, allowing access that's over 10 times faster than disk. And disk I/O bottlenecks become a thing of the past. I/O Express can go a long way in getting more out of your existing hardware, and can in fact help you delay costly hardware purchases. Now, those reasons stated, here are some that might not be so obvious:
- The budget's tight, and purchases are hard to come by. "Excuse me?" you might ask. "Last I checked, this is a reason NOT to buy."The reasoning is simple: I/O Express makes your system more efficient. If your budget is tight, you're going to need to squeeze every ounce of performance out of that system. Bear in mind this makes your jobs and applications more efficient, too. This makes your existing resources worth more. As in reason number 3, it will take your current hardware a lot further than it might have gone otherwise. Software costs a lot less than hardware. If you are in a restricted-budget situation, and you can get more out of the hardware you already have, your cost-justification is right there.
- You'll be moving off the OpenVMS platform in the near future. "What?" you might ask, aghast. "Now you're being ridiculous." First, it's been our experience that, when plans are made to phase a system out, it's usually about twice as long as the company thinks before the system is actually phased out. So, you may have that machine or those machines around longer than you think. Second, that system is going to be used while you've got it. You'll need to get as much as possible out of that machine or those machines. Handling disk I/O bottlenecks is a vital step in that direction, as detailed in reasons 3 and 4 above
- Based on system activity, applications, or other reasons, you've decided you don't need I/O Express. "Now you've completely lost it!" You say. "How could that be a reason to buy?" Well, you're right, it isn't. But we highly recommend that, before you make such a rash decision, at least try the product. We offer free trial demos. Many people have been very surprised to find out how badly disk I/O bottlenecks were affecting their performance, and how much I/O Express was able to help.
|
CONCLUSION
The bottom line is, I/O Express 4.3 can drastically improve performance and extend the life of your system. For the short term, or for the long term, it's worth it. Call +44 (0)1342 327477 and order today!
I/O Express Pricing
Our goal is to ensure you get the maximum performance possible for the money you spend. For this reason, we are currently offering the best pricing we've ever offered on I/O Express - including site licences in many instances.
For a quote on your OpenVMS machines, call:+44 (0)1342 327477
Let us help you licence I/O Express on all your OpenVMS machines!
Now you and your users can experience the I/O Express 4.3 performance difference for yourselves! We are now offering a free 30-day trial of I/O Express - a full working version of the software.
Call: +44 (0)1342 327477 and take advantage today!
|
