This is a discussion on Sizing replacement servers within the Sun Solaris Administration forums, part of the Solaris Operating System category; --> Hello all; I have a question regarding replacing older Sun servers, particularly regarding CPU horsepower. Here's a for instance: ...
| |||||||
| FAQ | Members List | Calendar | Search | Today's Posts | Mark Forums Read |
| ||||
| Hello all; I have a question regarding replacing older Sun servers, particularly regarding CPU horsepower. Here's a for instance: We have a couple of 420R servers, with 4 GB Ram and 4 X 450MHz CPUs. In spec-ing replacements, I'm looking at V440 servers. Since RAM is (relatively) cheap, I'm gonna spec the same amount of RAM, but CPUs are a different story. I can simply spec them with 4 CPUs, but on this model, the CPUs range from 1 GHz to 1.5 GHz. Plus, the price jump from 2 to 4 CPUs is about 10 grand (list, of course). Can I simply say "well, 4 X 450 GHz is 1800 GHz, so I'll spec them with 2 X 1.5 GHz CPUs for a total of 3 GHz of CPU power, and we'll be set for the current load, and have room for growth before we have to add processors"? I know, I know, I should have some stats gathered to gauge how well our 4 CPUs are doing for us now, but this was sprung on me last minute, and I need to get rough numbers to management ASAP. Any assistance, rules of thumb, or other advice gratefully acknowledged, as always. Joe D. |
| |||
| In article <1156965314.263568.257990@b28g2000cwb.googlegroups .com>, "Joe D." <newbie_from_newbie@yahoo.com> wrote: > Hello all; I have a question regarding replacing older Sun servers, > particularly regarding CPU horsepower. > > Here's a for instance: We have a couple of 420R servers, with 4 GB Ram > and 4 X 450MHz CPUs. In spec-ing replacements, I'm looking at V440 > servers. Since RAM is (relatively) cheap, I'm gonna spec the same > amount of RAM, but CPUs are a different story. I can simply spec them > with 4 CPUs, but on this model, the CPUs range from 1 GHz to 1.5 GHz. > Plus, the price jump from 2 to 4 CPUs is about 10 grand (list, of > course). > > Can I simply say "well, 4 X 450 GHz is 1800 GHz, so I'll spec them with > 2 X 1.5 GHz CPUs for a total of 3 GHz of CPU power, and we'll be set > for the current load, and have room for growth before we have to add > processors"? > > I know, I know, I should have some stats gathered to gauge how well our > 4 CPUs are doing for us now, but this was sprung on me last minute, and > I need to get rough numbers to management ASAP. > > Any assistance, rules of thumb, or other advice gratefully > acknowledged, as always. > > Joe D. Well, you could do "first blush" estimates giving as much time to it as the PHB who asked for it did. If you know what the budget is, get as much hardware as you can. Or quote the full memory and all 4 1.5GHz CPUs. In another, price full memory and 4 1GHz CPUs. In another, price full memory and 4 1.5GHz CPUs. Bet they go for the cheapest. If you think that will be the case, don't quote anything less than 4 CPUs until you know your workload doesn't need all of them. So what if they're faster. IMO, workloads don't scale the way you're thinking. Unless you have reviewed your sar and mpstat statistics and know that you're underutilizing the 4 CPUs, you'll probably miss them if you don't have them. You know your workload best. Work the system to get as much machine as you can. Maybe this is just a chinese firedrill and the PTB (powers that be) look at how much and tell you "Never mind". -- DeeDee, don't press that button! DeeDee! NO! Dee... |
| |||
| Joe D. wrote: > Hello all; I have a question regarding replacing older Sun servers, > particularly regarding CPU horsepower. > > Here's a for instance: We have a couple of 420R servers, with 4 GB Ram > and 4 X 450MHz CPUs. In spec-ing replacements, I'm looking at V440 > servers. Since RAM is (relatively) cheap, I'm gonna spec the same > amount of RAM, but CPUs are a different story. I can simply spec them > with 4 CPUs, but on this model, the CPUs range from 1 GHz to 1.5 GHz. > Plus, the price jump from 2 to 4 CPUs is about 10 grand (list, of > course). > > Can I simply say "well, 4 X 450 GHz is 1800 GHz, so I'll spec them with > 2 X 1.5 GHz CPUs for a total of 3 GHz of CPU power, and we'll be set > for the current load, and have room for growth before we have to add > processors"? > > I know, I know, I should have some stats gathered to gauge how well our > 4 CPUs are doing for us now, but this was sprung on me last minute, and > I need to get rough numbers to management ASAP. > > Any assistance, rules of thumb, or other advice gratefully > acknowledged, as always. > > Joe D. Today you need to look at the workload first of all. if the workload is transaction based in many threads go for the T1000/T2000 servers. If the workload is more HPC simulation with few threads get the fastest CPU's available. If you S/W is SPARC based that would be the new US-IV+ CPU's and the E2900 chassie. or possible the V440 with the 1.59 GHz CPU or the Ultra 45 workstation with the same CPU. To run a web/Apllication server / medium database on anything else than a T1000/T2000 today is daft, and a waste of time and money. //Lars |
| |||
| Joe D. wrote: > Hello all; I have a question regarding replacing older Sun servers, > particularly regarding CPU horsepower. > > Here's a for instance: We have a couple of 420R servers, with 4 GB Ram > and 4 X 450MHz CPUs. In spec-ing replacements, I'm looking at V440 > servers. Since RAM is (relatively) cheap, I'm gonna spec the same > amount of RAM, but CPUs are a different story. I can simply spec them > with 4 CPUs, but on this model, the CPUs range from 1 GHz to 1.5 GHz. > Plus, the price jump from 2 to 4 CPUs is about 10 grand (list, of > course). > > Can I simply say "well, 4 X 450 GHz is 1800 GHz, so I'll spec them with > 2 X 1.5 GHz CPUs for a total of 3 GHz of CPU power, and we'll be set > for the current load, and have room for growth before we have to add > processors"? > > I know, I know, I should have some stats gathered to gauge how well our > 4 CPUs are doing for us now, but this was sprung on me last minute, and > I need to get rough numbers to management ASAP. > > Any assistance, rules of thumb, or other advice gratefully > acknowledged, as always. > > Joe D. > 4 X CPUS does not compare well with N CPUS of different power. If there are not N schedulable processes (or threads), you can't use all N CPUs. I'd say your best bet is to tell management that they can have it done right or they can have it Tuesday. If you don't have a good picture of your workload and how your machines handle it, any recommendation you make stands a good chance of being useless (or worse). How much swapping are the current machines doing? If it's significant, get more memory. Are you actually keeping 4 CPUs busy? If not, it's a waste of money to configure four. If you are, you may want to think more or faster or both. Buying a new server is usually a substantial investment. Getting it right is, therefore, usually worth the time and effort. |
| |||
| On Wed, 30 Aug 2006, Joe D. wrote: > Hello all; I have a question regarding replacing older Sun servers, > particularly regarding CPU horsepower. Can you name the applications running on each server at the moment, or detail the business functions of each server? Also why are you replacing the servers? -- Chris |
| |||
| Chris Lawrence wrote: > Can you name the applications running on each server at the moment, or > detail the business functions of each server? Also why are you > replacing the servers? > Yes, I can; however, we are merely at the "let's see if we can get money allocated" stage right now. As several previous replies had pointed out (and I think your question was leading to as well), I will need to spend some time actually monitoring the load on each server before spec'ing something. So it's more of a spin the wheels exercise. One thing of note was a comment one of our VARs mentioned when I asked about sizing today's CPUs vs. yesterday's. Apparently, there is some sort of service that SUN provides that will allow you to calculate an 'm-value' (I may have mis-heard him) for processors, whereby they can tell you, all things being equal, the number and speed of processors needed to replace an existing configuration of X number of Y speed CPUs. I suspect that this would be a very subjective thing, and not reflective at all of some of the points made earlier (curent CPU load, type and amount of work being done, etc.). Has anyone heard of this 'service', or is this smoke and mirrors by Sun Services to drum up more business? Anyay; thanks as usuall to all who responded. As usual, I have gleaned more from this group than I am contributing, an imbalance I am hoping to rectify, just as soon as I see a posting that I can actually contribute something of worth to! : ) |
| |||
| Joe D. wrote: > Chris Lawrence wrote: > >>Can you name the applications running on each server at the moment, or >>detail the business functions of each server? Also why are you >>replacing the servers? >> > > > > Yes, I can; however, we are merely at the "let's see if we can get > money allocated" stage right now. As several previous replies had > pointed out (and I think your question was leading to as well), I will > need to spend some time actually monitoring the load on each server > before spec'ing something. So it's more of a spin the wheels exercise. > > One thing of note was a comment one of our VARs mentioned when I asked > about sizing today's CPUs vs. yesterday's. Apparently, there is some > sort of service that SUN provides that will allow you to calculate an > 'm-value' (I may have mis-heard him) for processors, whereby they can > tell you, all things being equal, the number and speed of processors > needed to replace an existing configuration of X number of Y speed > CPUs. > > I suspect that this would be a very subjective thing, and not > reflective at all of some of the points made earlier (curent CPU load, > type and amount of work being done, etc.). Has anyone heard of this > 'service', or is this smoke and mirrors by Sun Services to drum up more > business? Any, ANY, such thing has a large component of subjectiveness and guesswork. The only certain way to determine how your applications would perform on a given system is to benchmark them on that system. Sometimes that's easy to do. Sometimes a realistic benchmark would require installing 200 users and a quarter of a terrabyte of database and it's anything but easy. Having said that, sometimes an expert can make a very good guess; especially if said expert has no financial interest in the outcome. |
| ||||
| In article <0b-dnWYf-NKcSGDZnZ2dnUVZ_oqdnZ2d@comcast.com>, "Richard B. Gilbert" <rgilbert88@comcast.net> wrote: > Joe D. wrote: > > > Chris Lawrence wrote: > > > >>Can you name the applications running on each server at the moment, or > >>detail the business functions of each server? Also why are you > >>replacing the servers? > >> > > > > > > > > Yes, I can; however, we are merely at the "let's see if we can get > > money allocated" stage right now. As several previous replies had > > pointed out (and I think your question was leading to as well), I will > > need to spend some time actually monitoring the load on each server > > before spec'ing something. So it's more of a spin the wheels exercise. > > > > One thing of note was a comment one of our VARs mentioned when I asked > > about sizing today's CPUs vs. yesterday's. Apparently, there is some > > sort of service that SUN provides that will allow you to calculate an > > 'm-value' (I may have mis-heard him) for processors, whereby they can > > tell you, all things being equal, the number and speed of processors > > needed to replace an existing configuration of X number of Y speed > > CPUs. > > > > I suspect that this would be a very subjective thing, and not > > reflective at all of some of the points made earlier (curent CPU load, > > type and amount of work being done, etc.). Has anyone heard of this > > 'service', or is this smoke and mirrors by Sun Services to drum up more > > business? > > Any, ANY, such thing has a large component of subjectiveness and > guesswork. The only certain way to determine how your applications > would perform on a given system is to benchmark them on that system. > Sometimes that's easy to do. Sometimes a realistic benchmark would > require installing 200 users and a quarter of a terrabyte of database > and it's anything but easy. Having said that, sometimes an expert can > make a very good guess; especially if said expert has no financial > interest in the outcome. Recall the wise words of Brian Wong: "You can never say with certainty that a given configuration will be able to meet your needs. You can only be certain that a given configuration cannot meet your needs." Paraphrased of course. Josh |